1 /*
   2  * Copyright (c) 1999, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 // no precompiled headers
  26 #include "jvm.h"
  27 #include "classfile/classLoader.hpp"
  28 #include "classfile/systemDictionary.hpp"
  29 #include "classfile/vmSymbols.hpp"
  30 #include "code/icBuffer.hpp"
  31 #include "code/vtableStubs.hpp"
  32 #include "compiler/compileBroker.hpp"
  33 #include "compiler/disassembler.hpp"
  34 #include "interpreter/interpreter.hpp"
  35 #include "logging/log.hpp"
  36 #include "memory/allocation.inline.hpp"
  37 #include "memory/filemap.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "os_linux.inline.hpp"
  40 #include "os_share_linux.hpp"
  41 #include "osContainer_linux.hpp"
  42 #include "prims/jniFastGetField.hpp"
  43 #include "prims/jvm_misc.hpp"
  44 #include "runtime/arguments.hpp"
  45 #include "runtime/atomic.hpp"
  46 #include "runtime/extendedPC.hpp"
  47 #include "runtime/globals.hpp"
  48 #include "runtime/interfaceSupport.hpp"
  49 #include "runtime/init.hpp"
  50 #include "runtime/java.hpp"
  51 #include "runtime/javaCalls.hpp"
  52 #include "runtime/mutexLocker.hpp"
  53 #include "runtime/objectMonitor.hpp"
  54 #include "runtime/orderAccess.inline.hpp"
  55 #include "runtime/osThread.hpp"
  56 #include "runtime/perfMemory.hpp"
  57 #include "runtime/sharedRuntime.hpp"
  58 #include "runtime/statSampler.hpp"
  59 #include "runtime/stubRoutines.hpp"
  60 #include "runtime/thread.inline.hpp"
  61 #include "runtime/threadCritical.hpp"
  62 #include "runtime/timer.hpp"
  63 #include "semaphore_posix.hpp"
  64 #include "services/attachListener.hpp"
  65 #include "services/memTracker.hpp"
  66 #include "services/runtimeService.hpp"
  67 #include "utilities/align.hpp"
  68 #include "utilities/decoder.hpp"
  69 #include "utilities/defaultStream.hpp"
  70 #include "utilities/events.hpp"
  71 #include "utilities/elfFile.hpp"
  72 #include "utilities/growableArray.hpp"
  73 #include "utilities/macros.hpp"
  74 #include "utilities/vmError.hpp"
  75 
  76 // put OS-includes here
  77 # include <sys/types.h>
  78 # include <sys/mman.h>
  79 # include <sys/stat.h>
  80 # include <sys/select.h>
  81 # include <pthread.h>
  82 # include <signal.h>
  83 # include <errno.h>
  84 # include <dlfcn.h>
  85 # include <stdio.h>
  86 # include <unistd.h>
  87 # include <sys/resource.h>
  88 # include <pthread.h>
  89 # include <sys/stat.h>
  90 # include <sys/time.h>
  91 # include <sys/times.h>
  92 # include <sys/utsname.h>
  93 # include <sys/socket.h>
  94 # include <sys/wait.h>
  95 # include <pwd.h>
  96 # include <poll.h>
  97 # include <semaphore.h>
  98 # include <fcntl.h>
  99 # include <string.h>
 100 # include <syscall.h>
 101 # include <sys/sysinfo.h>
 102 # include <gnu/libc-version.h>
 103 # include <sys/ipc.h>
 104 # include <sys/shm.h>
 105 # include <link.h>
 106 # include <stdint.h>
 107 # include <inttypes.h>
 108 # include <sys/ioctl.h>
 109 
 110 #ifndef _GNU_SOURCE
 111   #define _GNU_SOURCE
 112   #include <sched.h>
 113   #undef _GNU_SOURCE
 114 #else
 115   #include <sched.h>
 116 #endif
 117 
 118 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
 119 // getrusage() is prepared to handle the associated failure.
 120 #ifndef RUSAGE_THREAD
 121   #define RUSAGE_THREAD   (1)               /* only the calling thread */
 122 #endif
 123 
 124 #define MAX_PATH    (2 * K)
 125 
 126 #define MAX_SECS 100000000
 127 
 128 // for timer info max values which include all bits
 129 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
 130 
 131 #define LARGEPAGES_BIT (1 << 6)
 132 ////////////////////////////////////////////////////////////////////////////////
 133 // global variables
 134 julong os::Linux::_physical_memory = 0;
 135 
 136 address   os::Linux::_initial_thread_stack_bottom = NULL;
 137 uintptr_t os::Linux::_initial_thread_stack_size   = 0;
 138 
 139 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
 140 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
 141 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
 142 Mutex* os::Linux::_createThread_lock = NULL;
 143 pthread_t os::Linux::_main_thread;
 144 int os::Linux::_page_size = -1;
 145 bool os::Linux::_supports_fast_thread_cpu_time = false;
 146 uint32_t os::Linux::_os_version = 0;
 147 const char * os::Linux::_glibc_version = NULL;
 148 const char * os::Linux::_libpthread_version = NULL;
 149 
 150 static jlong initial_time_count=0;
 151 
 152 static int clock_tics_per_sec = 100;
 153 
 154 // For diagnostics to print a message once. see run_periodic_checks
 155 static sigset_t check_signal_done;
 156 static bool check_signals = true;
 157 
 158 // Signal number used to suspend/resume a thread
 159 
 160 // do not use any signal number less than SIGSEGV, see 4355769
 161 static int SR_signum = SIGUSR2;
 162 sigset_t SR_sigset;
 163 
 164 // utility functions
 165 
 166 static int SR_initialize();
 167 
 168 julong os::available_memory() {
 169   return Linux::available_memory();
 170 }
 171 
 172 julong os::Linux::available_memory() {
 173   // values in struct sysinfo are "unsigned long"
 174   struct sysinfo si;
 175   julong avail_mem;
 176 
 177   if (OSContainer::is_containerized()) {
 178     jlong mem_limit, mem_usage;
 179     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 180       if ((mem_usage = OSContainer::memory_usage_in_bytes()) > 0) {
 181         if (mem_limit > mem_usage) {
 182           avail_mem = (julong)mem_limit - (julong)mem_usage;
 183         } else {
 184           avail_mem = 0;
 185         }
 186         log_trace(os)("available container memory: " JULONG_FORMAT, avail_mem);
 187         return avail_mem;
 188       } else {
 189         log_debug(os,container)("container memory usage call failed: " JLONG_FORMAT, mem_usage);
 190       }
 191     } else {
 192       log_debug(os,container)("container memory unlimited or failed: " JLONG_FORMAT, mem_limit);
 193     }
 194   }
 195 
 196   sysinfo(&si);
 197   avail_mem = (julong)si.freeram * si.mem_unit;
 198   log_trace(os)("available memory: " JULONG_FORMAT, avail_mem);
 199   return avail_mem;
 200 }
 201 
 202 julong os::physical_memory() {
 203   if (OSContainer::is_containerized()) {
 204     jlong mem_limit;
 205     if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
 206       log_trace(os)("total container memory: " JLONG_FORMAT, mem_limit);
 207       return (julong)mem_limit;
 208     } else {
 209       if (mem_limit == OSCONTAINER_ERROR) {
 210         log_debug(os,container)("container memory limit call failed");
 211       }
 212       if (mem_limit == -1) {
 213         log_debug(os,container)("container memory unlimited, using host value");
 214       }
 215     }
 216   }
 217 
 218   jlong phys_mem = Linux::physical_memory();
 219   log_trace(os)("total system memory: " JLONG_FORMAT, phys_mem);
 220   return phys_mem;
 221 }
 222 
 223 // Return true if user is running as root.
 224 
 225 bool os::have_special_privileges() {
 226   static bool init = false;
 227   static bool privileges = false;
 228   if (!init) {
 229     privileges = (getuid() != geteuid()) || (getgid() != getegid());
 230     init = true;
 231   }
 232   return privileges;
 233 }
 234 
 235 
 236 #ifndef SYS_gettid
 237 // i386: 224, ia64: 1105, amd64: 186, sparc 143
 238   #ifdef __ia64__
 239     #define SYS_gettid 1105
 240   #else
 241     #ifdef __i386__
 242       #define SYS_gettid 224
 243     #else
 244       #ifdef __amd64__
 245         #define SYS_gettid 186
 246       #else
 247         #ifdef __sparc__
 248           #define SYS_gettid 143
 249         #else
 250           #error define gettid for the arch
 251         #endif
 252       #endif
 253     #endif
 254   #endif
 255 #endif
 256 
 257 
 258 // pid_t gettid()
 259 //
 260 // Returns the kernel thread id of the currently running thread. Kernel
 261 // thread id is used to access /proc.
 262 pid_t os::Linux::gettid() {
 263   int rslt = syscall(SYS_gettid);
 264   assert(rslt != -1, "must be."); // old linuxthreads implementation?
 265   return (pid_t)rslt;
 266 }
 267 
 268 // Most versions of linux have a bug where the number of processors are
 269 // determined by looking at the /proc file system.  In a chroot environment,
 270 // the system call returns 1.  This causes the VM to act as if it is
 271 // a single processor and elide locking (see is_MP() call).
 272 static bool unsafe_chroot_detected = false;
 273 static const char *unstable_chroot_error = "/proc file system not found.\n"
 274                      "Java may be unstable running multithreaded in a chroot "
 275                      "environment on Linux when /proc filesystem is not mounted.";
 276 
 277 void os::Linux::initialize_system_info() {
 278   set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
 279   if (processor_count() == 1) {
 280     pid_t pid = os::Linux::gettid();
 281     char fname[32];
 282     jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
 283     FILE *fp = fopen(fname, "r");
 284     if (fp == NULL) {
 285       unsafe_chroot_detected = true;
 286     } else {
 287       fclose(fp);
 288     }
 289   }
 290   _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
 291   assert(processor_count() > 0, "linux error");
 292 }
 293 
 294 void os::init_system_properties_values() {
 295   // The next steps are taken in the product version:
 296   //
 297   // Obtain the JAVA_HOME value from the location of libjvm.so.
 298   // This library should be located at:
 299   // <JAVA_HOME>/lib/{client|server}/libjvm.so.
 300   //
 301   // If "/jre/lib/" appears at the right place in the path, then we
 302   // assume libjvm.so is installed in a JDK and we use this path.
 303   //
 304   // Otherwise exit with message: "Could not create the Java virtual machine."
 305   //
 306   // The following extra steps are taken in the debugging version:
 307   //
 308   // If "/jre/lib/" does NOT appear at the right place in the path
 309   // instead of exit check for $JAVA_HOME environment variable.
 310   //
 311   // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
 312   // then we append a fake suffix "hotspot/libjvm.so" to this path so
 313   // it looks like libjvm.so is installed there
 314   // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
 315   //
 316   // Otherwise exit.
 317   //
 318   // Important note: if the location of libjvm.so changes this
 319   // code needs to be changed accordingly.
 320 
 321   // See ld(1):
 322   //      The linker uses the following search paths to locate required
 323   //      shared libraries:
 324   //        1: ...
 325   //        ...
 326   //        7: The default directories, normally /lib and /usr/lib.
 327 #if defined(AMD64) || (defined(_LP64) && defined(SPARC)) || defined(PPC64) || defined(S390)
 328   #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
 329 #else
 330   #define DEFAULT_LIBPATH "/lib:/usr/lib"
 331 #endif
 332 
 333 // Base path of extensions installed on the system.
 334 #define SYS_EXT_DIR     "/usr/java/packages"
 335 #define EXTENSIONS_DIR  "/lib/ext"
 336 
 337   // Buffer that fits several sprintfs.
 338   // Note that the space for the colon and the trailing null are provided
 339   // by the nulls included by the sizeof operator.
 340   const size_t bufsize =
 341     MAX2((size_t)MAXPATHLEN,  // For dll_dir & friends.
 342          (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
 343   char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
 344 
 345   // sysclasspath, java_home, dll_dir
 346   {
 347     char *pslash;
 348     os::jvm_path(buf, bufsize);
 349 
 350     // Found the full path to libjvm.so.
 351     // Now cut the path to <java_home>/jre if we can.
 352     pslash = strrchr(buf, '/');
 353     if (pslash != NULL) {
 354       *pslash = '\0';            // Get rid of /libjvm.so.
 355     }
 356     pslash = strrchr(buf, '/');
 357     if (pslash != NULL) {
 358       *pslash = '\0';            // Get rid of /{client|server|hotspot}.
 359     }
 360     Arguments::set_dll_dir(buf);
 361 
 362     if (pslash != NULL) {
 363       pslash = strrchr(buf, '/');
 364       if (pslash != NULL) {
 365         *pslash = '\0';        // Get rid of /lib.
 366       }
 367     }
 368     Arguments::set_java_home(buf);
 369     set_boot_path('/', ':');
 370   }
 371 
 372   // Where to look for native libraries.
 373   //
 374   // Note: Due to a legacy implementation, most of the library path
 375   // is set in the launcher. This was to accomodate linking restrictions
 376   // on legacy Linux implementations (which are no longer supported).
 377   // Eventually, all the library path setting will be done here.
 378   //
 379   // However, to prevent the proliferation of improperly built native
 380   // libraries, the new path component /usr/java/packages is added here.
 381   // Eventually, all the library path setting will be done here.
 382   {
 383     // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
 384     // should always exist (until the legacy problem cited above is
 385     // addressed).
 386     const char *v = ::getenv("LD_LIBRARY_PATH");
 387     const char *v_colon = ":";
 388     if (v == NULL) { v = ""; v_colon = ""; }
 389     // That's +1 for the colon and +1 for the trailing '\0'.
 390     char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
 391                                                      strlen(v) + 1 +
 392                                                      sizeof(SYS_EXT_DIR) + sizeof("/lib/") + sizeof(DEFAULT_LIBPATH) + 1,
 393                                                      mtInternal);
 394     sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib:" DEFAULT_LIBPATH, v, v_colon);
 395     Arguments::set_library_path(ld_library_path);
 396     FREE_C_HEAP_ARRAY(char, ld_library_path);
 397   }
 398 
 399   // Extensions directories.
 400   sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
 401   Arguments::set_ext_dirs(buf);
 402 
 403   FREE_C_HEAP_ARRAY(char, buf);
 404 
 405 #undef DEFAULT_LIBPATH
 406 #undef SYS_EXT_DIR
 407 #undef EXTENSIONS_DIR
 408 }
 409 
 410 ////////////////////////////////////////////////////////////////////////////////
 411 // breakpoint support
 412 
 413 void os::breakpoint() {
 414   BREAKPOINT;
 415 }
 416 
 417 extern "C" void breakpoint() {
 418   // use debugger to set breakpoint here
 419 }
 420 
 421 ////////////////////////////////////////////////////////////////////////////////
 422 // signal support
 423 
 424 debug_only(static bool signal_sets_initialized = false);
 425 static sigset_t unblocked_sigs, vm_sigs;
 426 
 427 bool os::Linux::is_sig_ignored(int sig) {
 428   struct sigaction oact;
 429   sigaction(sig, (struct sigaction*)NULL, &oact);
 430   void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*,  oact.sa_sigaction)
 431                                  : CAST_FROM_FN_PTR(void*,  oact.sa_handler);
 432   if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
 433     return true;
 434   } else {
 435     return false;
 436   }
 437 }
 438 
 439 void os::Linux::signal_sets_init() {
 440   // Should also have an assertion stating we are still single-threaded.
 441   assert(!signal_sets_initialized, "Already initialized");
 442   // Fill in signals that are necessarily unblocked for all threads in
 443   // the VM. Currently, we unblock the following signals:
 444   // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
 445   //                         by -Xrs (=ReduceSignalUsage));
 446   // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
 447   // other threads. The "ReduceSignalUsage" boolean tells us not to alter
 448   // the dispositions or masks wrt these signals.
 449   // Programs embedding the VM that want to use the above signals for their
 450   // own purposes must, at this time, use the "-Xrs" option to prevent
 451   // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
 452   // (See bug 4345157, and other related bugs).
 453   // In reality, though, unblocking these signals is really a nop, since
 454   // these signals are not blocked by default.
 455   sigemptyset(&unblocked_sigs);
 456   sigaddset(&unblocked_sigs, SIGILL);
 457   sigaddset(&unblocked_sigs, SIGSEGV);
 458   sigaddset(&unblocked_sigs, SIGBUS);
 459   sigaddset(&unblocked_sigs, SIGFPE);
 460 #if defined(PPC64)
 461   sigaddset(&unblocked_sigs, SIGTRAP);
 462 #endif
 463   sigaddset(&unblocked_sigs, SR_signum);
 464 
 465   if (!ReduceSignalUsage) {
 466     if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
 467       sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
 468     }
 469     if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
 470       sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
 471     }
 472     if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
 473       sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
 474     }
 475   }
 476   // Fill in signals that are blocked by all but the VM thread.
 477   sigemptyset(&vm_sigs);
 478   if (!ReduceSignalUsage) {
 479     sigaddset(&vm_sigs, BREAK_SIGNAL);
 480   }
 481   debug_only(signal_sets_initialized = true);
 482 
 483 }
 484 
 485 // These are signals that are unblocked while a thread is running Java.
 486 // (For some reason, they get blocked by default.)
 487 sigset_t* os::Linux::unblocked_signals() {
 488   assert(signal_sets_initialized, "Not initialized");
 489   return &unblocked_sigs;
 490 }
 491 
 492 // These are the signals that are blocked while a (non-VM) thread is
 493 // running Java. Only the VM thread handles these signals.
 494 sigset_t* os::Linux::vm_signals() {
 495   assert(signal_sets_initialized, "Not initialized");
 496   return &vm_sigs;
 497 }
 498 
 499 void os::Linux::hotspot_sigmask(Thread* thread) {
 500 
 501   //Save caller's signal mask before setting VM signal mask
 502   sigset_t caller_sigmask;
 503   pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
 504 
 505   OSThread* osthread = thread->osthread();
 506   osthread->set_caller_sigmask(caller_sigmask);
 507 
 508   pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
 509 
 510   if (!ReduceSignalUsage) {
 511     if (thread->is_VM_thread()) {
 512       // Only the VM thread handles BREAK_SIGNAL ...
 513       pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
 514     } else {
 515       // ... all other threads block BREAK_SIGNAL
 516       pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
 517     }
 518   }
 519 }
 520 
 521 //////////////////////////////////////////////////////////////////////////////
 522 // detecting pthread library
 523 
 524 void os::Linux::libpthread_init() {
 525   // Save glibc and pthread version strings.
 526 #if !defined(_CS_GNU_LIBC_VERSION) || \
 527     !defined(_CS_GNU_LIBPTHREAD_VERSION)
 528   #error "glibc too old (< 2.3.2)"
 529 #endif
 530 
 531   size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
 532   assert(n > 0, "cannot retrieve glibc version");
 533   char *str = (char *)malloc(n, mtInternal);
 534   confstr(_CS_GNU_LIBC_VERSION, str, n);
 535   os::Linux::set_glibc_version(str);
 536 
 537   n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
 538   assert(n > 0, "cannot retrieve pthread version");
 539   str = (char *)malloc(n, mtInternal);
 540   confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
 541   os::Linux::set_libpthread_version(str);
 542 }
 543 
 544 /////////////////////////////////////////////////////////////////////////////
 545 // thread stack expansion
 546 
 547 // os::Linux::manually_expand_stack() takes care of expanding the thread
 548 // stack. Note that this is normally not needed: pthread stacks allocate
 549 // thread stack using mmap() without MAP_NORESERVE, so the stack is already
 550 // committed. Therefore it is not necessary to expand the stack manually.
 551 //
 552 // Manually expanding the stack was historically needed on LinuxThreads
 553 // thread stacks, which were allocated with mmap(MAP_GROWSDOWN). Nowadays
 554 // it is kept to deal with very rare corner cases:
 555 //
 556 // For one, user may run the VM on an own implementation of threads
 557 // whose stacks are - like the old LinuxThreads - implemented using
 558 // mmap(MAP_GROWSDOWN).
 559 //
 560 // Also, this coding may be needed if the VM is running on the primordial
 561 // thread. Normally we avoid running on the primordial thread; however,
 562 // user may still invoke the VM on the primordial thread.
 563 //
 564 // The following historical comment describes the details about running
 565 // on a thread stack allocated with mmap(MAP_GROWSDOWN):
 566 
 567 
 568 // Force Linux kernel to expand current thread stack. If "bottom" is close
 569 // to the stack guard, caller should block all signals.
 570 //
 571 // MAP_GROWSDOWN:
 572 //   A special mmap() flag that is used to implement thread stacks. It tells
 573 //   kernel that the memory region should extend downwards when needed. This
 574 //   allows early versions of LinuxThreads to only mmap the first few pages
 575 //   when creating a new thread. Linux kernel will automatically expand thread
 576 //   stack as needed (on page faults).
 577 //
 578 //   However, because the memory region of a MAP_GROWSDOWN stack can grow on
 579 //   demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
 580 //   region, it's hard to tell if the fault is due to a legitimate stack
 581 //   access or because of reading/writing non-exist memory (e.g. buffer
 582 //   overrun). As a rule, if the fault happens below current stack pointer,
 583 //   Linux kernel does not expand stack, instead a SIGSEGV is sent to the
 584 //   application (see Linux kernel fault.c).
 585 //
 586 //   This Linux feature can cause SIGSEGV when VM bangs thread stack for
 587 //   stack overflow detection.
 588 //
 589 //   Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
 590 //   not use MAP_GROWSDOWN.
 591 //
 592 // To get around the problem and allow stack banging on Linux, we need to
 593 // manually expand thread stack after receiving the SIGSEGV.
 594 //
 595 // There are two ways to expand thread stack to address "bottom", we used
 596 // both of them in JVM before 1.5:
 597 //   1. adjust stack pointer first so that it is below "bottom", and then
 598 //      touch "bottom"
 599 //   2. mmap() the page in question
 600 //
 601 // Now alternate signal stack is gone, it's harder to use 2. For instance,
 602 // if current sp is already near the lower end of page 101, and we need to
 603 // call mmap() to map page 100, it is possible that part of the mmap() frame
 604 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
 605 // That will destroy the mmap() frame and cause VM to crash.
 606 //
 607 // The following code works by adjusting sp first, then accessing the "bottom"
 608 // page to force a page fault. Linux kernel will then automatically expand the
 609 // stack mapping.
 610 //
 611 // _expand_stack_to() assumes its frame size is less than page size, which
 612 // should always be true if the function is not inlined.
 613 
 614 static void NOINLINE _expand_stack_to(address bottom) {
 615   address sp;
 616   size_t size;
 617   volatile char *p;
 618 
 619   // Adjust bottom to point to the largest address within the same page, it
 620   // gives us a one-page buffer if alloca() allocates slightly more memory.
 621   bottom = (address)align_down((uintptr_t)bottom, os::Linux::page_size());
 622   bottom += os::Linux::page_size() - 1;
 623 
 624   // sp might be slightly above current stack pointer; if that's the case, we
 625   // will alloca() a little more space than necessary, which is OK. Don't use
 626   // os::current_stack_pointer(), as its result can be slightly below current
 627   // stack pointer, causing us to not alloca enough to reach "bottom".
 628   sp = (address)&sp;
 629 
 630   if (sp > bottom) {
 631     size = sp - bottom;
 632     p = (volatile char *)alloca(size);
 633     assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
 634     p[0] = '\0';
 635   }
 636 }
 637 
 638 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
 639   assert(t!=NULL, "just checking");
 640   assert(t->osthread()->expanding_stack(), "expand should be set");
 641   assert(t->stack_base() != NULL, "stack_base was not initialized");
 642 
 643   if (addr <  t->stack_base() && addr >= t->stack_reserved_zone_base()) {
 644     sigset_t mask_all, old_sigset;
 645     sigfillset(&mask_all);
 646     pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
 647     _expand_stack_to(addr);
 648     pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
 649     return true;
 650   }
 651   return false;
 652 }
 653 
 654 //////////////////////////////////////////////////////////////////////////////
 655 // create new thread
 656 
 657 // Thread start routine for all newly created threads
 658 static void *thread_native_entry(Thread *thread) {
 659   // Try to randomize the cache line index of hot stack frames.
 660   // This helps when threads of the same stack traces evict each other's
 661   // cache lines. The threads can be either from the same JVM instance, or
 662   // from different JVM instances. The benefit is especially true for
 663   // processors with hyperthreading technology.
 664   static int counter = 0;
 665   int pid = os::current_process_id();
 666   alloca(((pid ^ counter++) & 7) * 128);
 667 
 668   thread->initialize_thread_current();
 669 
 670   OSThread* osthread = thread->osthread();
 671   Monitor* sync = osthread->startThread_lock();
 672 
 673   osthread->set_thread_id(os::current_thread_id());
 674 
 675   log_info(os, thread)("Thread is alive (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 676     os::current_thread_id(), (uintx) pthread_self());
 677 
 678   if (UseNUMA) {
 679     int lgrp_id = os::numa_get_group_id();
 680     if (lgrp_id != -1) {
 681       thread->set_lgrp_id(lgrp_id);
 682     }
 683   }
 684   // initialize signal mask for this thread
 685   os::Linux::hotspot_sigmask(thread);
 686 
 687   // initialize floating point control register
 688   os::Linux::init_thread_fpu_state();
 689 
 690   // handshaking with parent thread
 691   {
 692     MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
 693 
 694     // notify parent thread
 695     osthread->set_state(INITIALIZED);
 696     sync->notify_all();
 697 
 698     // wait until os::start_thread()
 699     while (osthread->get_state() == INITIALIZED) {
 700       sync->wait(Mutex::_no_safepoint_check_flag);
 701     }
 702   }
 703 
 704   // call one more level start routine
 705   thread->run();
 706 
 707   log_info(os, thread)("Thread finished (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 708     os::current_thread_id(), (uintx) pthread_self());
 709 
 710   // If a thread has not deleted itself ("delete this") as part of its
 711   // termination sequence, we have to ensure thread-local-storage is
 712   // cleared before we actually terminate. No threads should ever be
 713   // deleted asynchronously with respect to their termination.
 714   if (Thread::current_or_null_safe() != NULL) {
 715     assert(Thread::current_or_null_safe() == thread, "current thread is wrong");
 716     thread->clear_thread_current();
 717   }
 718 
 719   return 0;
 720 }
 721 
 722 bool os::create_thread(Thread* thread, ThreadType thr_type,
 723                        size_t req_stack_size) {
 724   assert(thread->osthread() == NULL, "caller responsible");
 725 
 726   // Allocate the OSThread object
 727   OSThread* osthread = new OSThread(NULL, NULL);
 728   if (osthread == NULL) {
 729     return false;
 730   }
 731 
 732   // set the correct thread state
 733   osthread->set_thread_type(thr_type);
 734 
 735   // Initial state is ALLOCATED but not INITIALIZED
 736   osthread->set_state(ALLOCATED);
 737 
 738   thread->set_osthread(osthread);
 739 
 740   // init thread attributes
 741   pthread_attr_t attr;
 742   pthread_attr_init(&attr);
 743   pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 744 
 745   // Calculate stack size if it's not specified by caller.
 746   size_t stack_size = os::Posix::get_initial_stack_size(thr_type, req_stack_size);
 747   // In the Linux NPTL pthread implementation the guard size mechanism
 748   // is not implemented properly. The posix standard requires adding
 749   // the size of the guard pages to the stack size, instead Linux
 750   // takes the space out of 'stacksize'. Thus we adapt the requested
 751   // stack_size by the size of the guard pages to mimick proper
 752   // behaviour. However, be careful not to end up with a size
 753   // of zero due to overflow. Don't add the guard page in that case.
 754   size_t guard_size = os::Linux::default_guard_size(thr_type);
 755   if (stack_size <= SIZE_MAX - guard_size) {
 756     stack_size += guard_size;
 757   }
 758   assert(is_aligned(stack_size, os::vm_page_size()), "stack_size not aligned");
 759 
 760   int status = pthread_attr_setstacksize(&attr, stack_size);
 761   assert_status(status == 0, status, "pthread_attr_setstacksize");
 762 
 763   // Configure glibc guard page.
 764   pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
 765 
 766   ThreadState state;
 767 
 768   {
 769     pthread_t tid;
 770     int ret = pthread_create(&tid, &attr, (void* (*)(void*)) thread_native_entry, thread);
 771 
 772     char buf[64];
 773     if (ret == 0) {
 774       log_info(os, thread)("Thread started (pthread id: " UINTX_FORMAT ", attributes: %s). ",
 775         (uintx) tid, os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 776     } else {
 777       log_warning(os, thread)("Failed to start thread - pthread_create failed (%s) for attributes: %s.",
 778         os::errno_name(ret), os::Posix::describe_pthread_attr(buf, sizeof(buf), &attr));
 779     }
 780 
 781     pthread_attr_destroy(&attr);
 782 
 783     if (ret != 0) {
 784       // Need to clean up stuff we've allocated so far
 785       thread->set_osthread(NULL);
 786       delete osthread;
 787       return false;
 788     }
 789 
 790     // Store pthread info into the OSThread
 791     osthread->set_pthread_id(tid);
 792 
 793     // Wait until child thread is either initialized or aborted
 794     {
 795       Monitor* sync_with_child = osthread->startThread_lock();
 796       MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 797       while ((state = osthread->get_state()) == ALLOCATED) {
 798         sync_with_child->wait(Mutex::_no_safepoint_check_flag);
 799       }
 800     }
 801   }
 802 
 803   // Aborted due to thread limit being reached
 804   if (state == ZOMBIE) {
 805     thread->set_osthread(NULL);
 806     delete osthread;
 807     return false;
 808   }
 809 
 810   // The thread is returned suspended (in state INITIALIZED),
 811   // and is started higher up in the call chain
 812   assert(state == INITIALIZED, "race condition");
 813   return true;
 814 }
 815 
 816 /////////////////////////////////////////////////////////////////////////////
 817 // attach existing thread
 818 
 819 // bootstrap the main thread
 820 bool os::create_main_thread(JavaThread* thread) {
 821   assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
 822   return create_attached_thread(thread);
 823 }
 824 
 825 bool os::create_attached_thread(JavaThread* thread) {
 826 #ifdef ASSERT
 827   thread->verify_not_published();
 828 #endif
 829 
 830   // Allocate the OSThread object
 831   OSThread* osthread = new OSThread(NULL, NULL);
 832 
 833   if (osthread == NULL) {
 834     return false;
 835   }
 836 
 837   // Store pthread info into the OSThread
 838   osthread->set_thread_id(os::Linux::gettid());
 839   osthread->set_pthread_id(::pthread_self());
 840 
 841   // initialize floating point control register
 842   os::Linux::init_thread_fpu_state();
 843 
 844   // Initial thread state is RUNNABLE
 845   osthread->set_state(RUNNABLE);
 846 
 847   thread->set_osthread(osthread);
 848 
 849   if (UseNUMA) {
 850     int lgrp_id = os::numa_get_group_id();
 851     if (lgrp_id != -1) {
 852       thread->set_lgrp_id(lgrp_id);
 853     }
 854   }
 855 
 856   if (os::Linux::is_initial_thread()) {
 857     // If current thread is initial thread, its stack is mapped on demand,
 858     // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
 859     // the entire stack region to avoid SEGV in stack banging.
 860     // It is also useful to get around the heap-stack-gap problem on SuSE
 861     // kernel (see 4821821 for details). We first expand stack to the top
 862     // of yellow zone, then enable stack yellow zone (order is significant,
 863     // enabling yellow zone first will crash JVM on SuSE Linux), so there
 864     // is no gap between the last two virtual memory regions.
 865 
 866     JavaThread *jt = (JavaThread *)thread;
 867     address addr = jt->stack_reserved_zone_base();
 868     assert(addr != NULL, "initialization problem?");
 869     assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
 870 
 871     osthread->set_expanding_stack();
 872     os::Linux::manually_expand_stack(jt, addr);
 873     osthread->clear_expanding_stack();
 874   }
 875 
 876   // initialize signal mask for this thread
 877   // and save the caller's signal mask
 878   os::Linux::hotspot_sigmask(thread);
 879 
 880   log_info(os, thread)("Thread attached (tid: " UINTX_FORMAT ", pthread id: " UINTX_FORMAT ").",
 881     os::current_thread_id(), (uintx) pthread_self());
 882 
 883   return true;
 884 }
 885 
 886 void os::pd_start_thread(Thread* thread) {
 887   OSThread * osthread = thread->osthread();
 888   assert(osthread->get_state() != INITIALIZED, "just checking");
 889   Monitor* sync_with_child = osthread->startThread_lock();
 890   MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
 891   sync_with_child->notify();
 892 }
 893 
 894 // Free Linux resources related to the OSThread
 895 void os::free_thread(OSThread* osthread) {
 896   assert(osthread != NULL, "osthread not set");
 897 
 898   // We are told to free resources of the argument thread,
 899   // but we can only really operate on the current thread.
 900   assert(Thread::current()->osthread() == osthread,
 901          "os::free_thread but not current thread");
 902 
 903 #ifdef ASSERT
 904   sigset_t current;
 905   sigemptyset(&current);
 906   pthread_sigmask(SIG_SETMASK, NULL, &current);
 907   assert(!sigismember(&current, SR_signum), "SR signal should not be blocked!");
 908 #endif
 909 
 910   // Restore caller's signal mask
 911   sigset_t sigmask = osthread->caller_sigmask();
 912   pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
 913 
 914   delete osthread;
 915 }
 916 
 917 //////////////////////////////////////////////////////////////////////////////
 918 // initial thread
 919 
 920 // Check if current thread is the initial thread, similar to Solaris thr_main.
 921 bool os::Linux::is_initial_thread(void) {
 922   char dummy;
 923   // If called before init complete, thread stack bottom will be null.
 924   // Can be called if fatal error occurs before initialization.
 925   if (initial_thread_stack_bottom() == NULL) return false;
 926   assert(initial_thread_stack_bottom() != NULL &&
 927          initial_thread_stack_size()   != 0,
 928          "os::init did not locate initial thread's stack region");
 929   if ((address)&dummy >= initial_thread_stack_bottom() &&
 930       (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
 931     return true;
 932   } else {
 933     return false;
 934   }
 935 }
 936 
 937 // Find the virtual memory area that contains addr
 938 static bool find_vma(address addr, address* vma_low, address* vma_high) {
 939   FILE *fp = fopen("/proc/self/maps", "r");
 940   if (fp) {
 941     address low, high;
 942     while (!feof(fp)) {
 943       if (fscanf(fp, "%p-%p", &low, &high) == 2) {
 944         if (low <= addr && addr < high) {
 945           if (vma_low)  *vma_low  = low;
 946           if (vma_high) *vma_high = high;
 947           fclose(fp);
 948           return true;
 949         }
 950       }
 951       for (;;) {
 952         int ch = fgetc(fp);
 953         if (ch == EOF || ch == (int)'\n') break;
 954       }
 955     }
 956     fclose(fp);
 957   }
 958   return false;
 959 }
 960 
 961 // Locate initial thread stack. This special handling of initial thread stack
 962 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
 963 // bogus value for the primordial process thread. While the launcher has created
 964 // the VM in a new thread since JDK 6, we still have to allow for the use of the
 965 // JNI invocation API from a primordial thread.
 966 void os::Linux::capture_initial_stack(size_t max_size) {
 967 
 968   // max_size is either 0 (which means accept OS default for thread stacks) or
 969   // a user-specified value known to be at least the minimum needed. If we
 970   // are actually on the primordial thread we can make it appear that we have a
 971   // smaller max_size stack by inserting the guard pages at that location. But we
 972   // cannot do anything to emulate a larger stack than what has been provided by
 973   // the OS or threading library. In fact if we try to use a stack greater than
 974   // what is set by rlimit then we will crash the hosting process.
 975 
 976   // Maximum stack size is the easy part, get it from RLIMIT_STACK.
 977   // If this is "unlimited" then it will be a huge value.
 978   struct rlimit rlim;
 979   getrlimit(RLIMIT_STACK, &rlim);
 980   size_t stack_size = rlim.rlim_cur;
 981 
 982   // 6308388: a bug in ld.so will relocate its own .data section to the
 983   //   lower end of primordial stack; reduce ulimit -s value a little bit
 984   //   so we won't install guard page on ld.so's data section.
 985   stack_size -= 2 * page_size();
 986 
 987   // Try to figure out where the stack base (top) is. This is harder.
 988   //
 989   // When an application is started, glibc saves the initial stack pointer in
 990   // a global variable "__libc_stack_end", which is then used by system
 991   // libraries. __libc_stack_end should be pretty close to stack top. The
 992   // variable is available since the very early days. However, because it is
 993   // a private interface, it could disappear in the future.
 994   //
 995   // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
 996   // to __libc_stack_end, it is very close to stack top, but isn't the real
 997   // stack top. Note that /proc may not exist if VM is running as a chroot
 998   // program, so reading /proc/<pid>/stat could fail. Also the contents of
 999   // /proc/<pid>/stat could change in the future (though unlikely).
1000   //
1001   // We try __libc_stack_end first. If that doesn't work, look for
1002   // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1003   // as a hint, which should work well in most cases.
1004 
1005   uintptr_t stack_start;
1006 
1007   // try __libc_stack_end first
1008   uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1009   if (p && *p) {
1010     stack_start = *p;
1011   } else {
1012     // see if we can get the start_stack field from /proc/self/stat
1013     FILE *fp;
1014     int pid;
1015     char state;
1016     int ppid;
1017     int pgrp;
1018     int session;
1019     int nr;
1020     int tpgrp;
1021     unsigned long flags;
1022     unsigned long minflt;
1023     unsigned long cminflt;
1024     unsigned long majflt;
1025     unsigned long cmajflt;
1026     unsigned long utime;
1027     unsigned long stime;
1028     long cutime;
1029     long cstime;
1030     long prio;
1031     long nice;
1032     long junk;
1033     long it_real;
1034     uintptr_t start;
1035     uintptr_t vsize;
1036     intptr_t rss;
1037     uintptr_t rsslim;
1038     uintptr_t scodes;
1039     uintptr_t ecode;
1040     int i;
1041 
1042     // Figure what the primordial thread stack base is. Code is inspired
1043     // by email from Hans Boehm. /proc/self/stat begins with current pid,
1044     // followed by command name surrounded by parentheses, state, etc.
1045     char stat[2048];
1046     int statlen;
1047 
1048     fp = fopen("/proc/self/stat", "r");
1049     if (fp) {
1050       statlen = fread(stat, 1, 2047, fp);
1051       stat[statlen] = '\0';
1052       fclose(fp);
1053 
1054       // Skip pid and the command string. Note that we could be dealing with
1055       // weird command names, e.g. user could decide to rename java launcher
1056       // to "java 1.4.2 :)", then the stat file would look like
1057       //                1234 (java 1.4.2 :)) R ... ...
1058       // We don't really need to know the command string, just find the last
1059       // occurrence of ")" and then start parsing from there. See bug 4726580.
1060       char * s = strrchr(stat, ')');
1061 
1062       i = 0;
1063       if (s) {
1064         // Skip blank chars
1065         do { s++; } while (s && isspace(*s));
1066 
1067 #define _UFM UINTX_FORMAT
1068 #define _DFM INTX_FORMAT
1069 
1070         //                                     1   1   1   1   1   1   1   1   1   1   2   2    2    2    2    2    2    2    2
1071         //              3  4  5  6  7  8   9   0   1   2   3   4   5   6   7   8   9   0   1    2    3    4    5    6    7    8
1072         i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1073                    &state,          // 3  %c
1074                    &ppid,           // 4  %d
1075                    &pgrp,           // 5  %d
1076                    &session,        // 6  %d
1077                    &nr,             // 7  %d
1078                    &tpgrp,          // 8  %d
1079                    &flags,          // 9  %lu
1080                    &minflt,         // 10 %lu
1081                    &cminflt,        // 11 %lu
1082                    &majflt,         // 12 %lu
1083                    &cmajflt,        // 13 %lu
1084                    &utime,          // 14 %lu
1085                    &stime,          // 15 %lu
1086                    &cutime,         // 16 %ld
1087                    &cstime,         // 17 %ld
1088                    &prio,           // 18 %ld
1089                    &nice,           // 19 %ld
1090                    &junk,           // 20 %ld
1091                    &it_real,        // 21 %ld
1092                    &start,          // 22 UINTX_FORMAT
1093                    &vsize,          // 23 UINTX_FORMAT
1094                    &rss,            // 24 INTX_FORMAT
1095                    &rsslim,         // 25 UINTX_FORMAT
1096                    &scodes,         // 26 UINTX_FORMAT
1097                    &ecode,          // 27 UINTX_FORMAT
1098                    &stack_start);   // 28 UINTX_FORMAT
1099       }
1100 
1101 #undef _UFM
1102 #undef _DFM
1103 
1104       if (i != 28 - 2) {
1105         assert(false, "Bad conversion from /proc/self/stat");
1106         // product mode - assume we are the initial thread, good luck in the
1107         // embedded case.
1108         warning("Can't detect initial thread stack location - bad conversion");
1109         stack_start = (uintptr_t) &rlim;
1110       }
1111     } else {
1112       // For some reason we can't open /proc/self/stat (for example, running on
1113       // FreeBSD with a Linux emulator, or inside chroot), this should work for
1114       // most cases, so don't abort:
1115       warning("Can't detect initial thread stack location - no /proc/self/stat");
1116       stack_start = (uintptr_t) &rlim;
1117     }
1118   }
1119 
1120   // Now we have a pointer (stack_start) very close to the stack top, the
1121   // next thing to do is to figure out the exact location of stack top. We
1122   // can find out the virtual memory area that contains stack_start by
1123   // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1124   // and its upper limit is the real stack top. (again, this would fail if
1125   // running inside chroot, because /proc may not exist.)
1126 
1127   uintptr_t stack_top;
1128   address low, high;
1129   if (find_vma((address)stack_start, &low, &high)) {
1130     // success, "high" is the true stack top. (ignore "low", because initial
1131     // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1132     stack_top = (uintptr_t)high;
1133   } else {
1134     // failed, likely because /proc/self/maps does not exist
1135     warning("Can't detect initial thread stack location - find_vma failed");
1136     // best effort: stack_start is normally within a few pages below the real
1137     // stack top, use it as stack top, and reduce stack size so we won't put
1138     // guard page outside stack.
1139     stack_top = stack_start;
1140     stack_size -= 16 * page_size();
1141   }
1142 
1143   // stack_top could be partially down the page so align it
1144   stack_top = align_up(stack_top, page_size());
1145 
1146   // Allowed stack value is minimum of max_size and what we derived from rlimit
1147   if (max_size > 0) {
1148     _initial_thread_stack_size = MIN2(max_size, stack_size);
1149   } else {
1150     // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1151     // clamp it at 8MB as we do on Solaris
1152     _initial_thread_stack_size = MIN2(stack_size, 8*M);
1153   }
1154   _initial_thread_stack_size = align_down(_initial_thread_stack_size, page_size());
1155   _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1156 
1157   assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1158 
1159   if (log_is_enabled(Info, os, thread)) {
1160     // See if we seem to be on primordial process thread
1161     bool primordial = uintptr_t(&rlim) > uintptr_t(_initial_thread_stack_bottom) &&
1162                       uintptr_t(&rlim) < stack_top;
1163 
1164     log_info(os, thread)("Capturing initial stack in %s thread: req. size: " SIZE_FORMAT "K, actual size: "
1165                          SIZE_FORMAT "K, top=" INTPTR_FORMAT ", bottom=" INTPTR_FORMAT,
1166                          primordial ? "primordial" : "user", max_size / K,  _initial_thread_stack_size / K,
1167                          stack_top, intptr_t(_initial_thread_stack_bottom));
1168   }
1169 }
1170 
1171 ////////////////////////////////////////////////////////////////////////////////
1172 // time support
1173 
1174 // Time since start-up in seconds to a fine granularity.
1175 // Used by VMSelfDestructTimer and the MemProfiler.
1176 double os::elapsedTime() {
1177 
1178   return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1179 }
1180 
1181 jlong os::elapsed_counter() {
1182   return javaTimeNanos() - initial_time_count;
1183 }
1184 
1185 jlong os::elapsed_frequency() {
1186   return NANOSECS_PER_SEC; // nanosecond resolution
1187 }
1188 
1189 bool os::supports_vtime() { return true; }
1190 bool os::enable_vtime()   { return false; }
1191 bool os::vtime_enabled()  { return false; }
1192 
1193 double os::elapsedVTime() {
1194   struct rusage usage;
1195   int retval = getrusage(RUSAGE_THREAD, &usage);
1196   if (retval == 0) {
1197     return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1198   } else {
1199     // better than nothing, but not much
1200     return elapsedTime();
1201   }
1202 }
1203 
1204 jlong os::javaTimeMillis() {
1205   timeval time;
1206   int status = gettimeofday(&time, NULL);
1207   assert(status != -1, "linux error");
1208   return jlong(time.tv_sec) * 1000  +  jlong(time.tv_usec / 1000);
1209 }
1210 
1211 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1212   timeval time;
1213   int status = gettimeofday(&time, NULL);
1214   assert(status != -1, "linux error");
1215   seconds = jlong(time.tv_sec);
1216   nanos = jlong(time.tv_usec) * 1000;
1217 }
1218 
1219 
1220 #ifndef CLOCK_MONOTONIC
1221   #define CLOCK_MONOTONIC (1)
1222 #endif
1223 
1224 void os::Linux::clock_init() {
1225   // we do dlopen's in this particular order due to bug in linux
1226   // dynamical loader (see 6348968) leading to crash on exit
1227   void* handle = dlopen("librt.so.1", RTLD_LAZY);
1228   if (handle == NULL) {
1229     handle = dlopen("librt.so", RTLD_LAZY);
1230   }
1231 
1232   if (handle) {
1233     int (*clock_getres_func)(clockid_t, struct timespec*) =
1234            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1235     int (*clock_gettime_func)(clockid_t, struct timespec*) =
1236            (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1237     if (clock_getres_func && clock_gettime_func) {
1238       // See if monotonic clock is supported by the kernel. Note that some
1239       // early implementations simply return kernel jiffies (updated every
1240       // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1241       // for nano time (though the monotonic property is still nice to have).
1242       // It's fixed in newer kernels, however clock_getres() still returns
1243       // 1/HZ. We check if clock_getres() works, but will ignore its reported
1244       // resolution for now. Hopefully as people move to new kernels, this
1245       // won't be a problem.
1246       struct timespec res;
1247       struct timespec tp;
1248       if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1249           clock_gettime_func(CLOCK_MONOTONIC, &tp)  == 0) {
1250         // yes, monotonic clock is supported
1251         _clock_gettime = clock_gettime_func;
1252         return;
1253       } else {
1254         // close librt if there is no monotonic clock
1255         dlclose(handle);
1256       }
1257     }
1258   }
1259   warning("No monotonic clock was available - timed services may " \
1260           "be adversely affected if the time-of-day clock changes");
1261 }
1262 
1263 #ifndef SYS_clock_getres
1264   #if defined(X86) || defined(PPC64) || defined(S390)
1265     #define SYS_clock_getres AMD64_ONLY(229) IA32_ONLY(266) PPC64_ONLY(247) S390_ONLY(261)
1266     #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1267   #else
1268     #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1269     #define sys_clock_getres(x,y)  -1
1270   #endif
1271 #else
1272   #define sys_clock_getres(x,y)  ::syscall(SYS_clock_getres, x, y)
1273 #endif
1274 
1275 void os::Linux::fast_thread_clock_init() {
1276   if (!UseLinuxPosixThreadCPUClocks) {
1277     return;
1278   }
1279   clockid_t clockid;
1280   struct timespec tp;
1281   int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1282       (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1283 
1284   // Switch to using fast clocks for thread cpu time if
1285   // the sys_clock_getres() returns 0 error code.
1286   // Note, that some kernels may support the current thread
1287   // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1288   // returned by the pthread_getcpuclockid().
1289   // If the fast Posix clocks are supported then the sys_clock_getres()
1290   // must return at least tp.tv_sec == 0 which means a resolution
1291   // better than 1 sec. This is extra check for reliability.
1292 
1293   if (pthread_getcpuclockid_func &&
1294       pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1295       sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1296     _supports_fast_thread_cpu_time = true;
1297     _pthread_getcpuclockid = pthread_getcpuclockid_func;
1298   }
1299 }
1300 
1301 jlong os::javaTimeNanos() {
1302   if (os::supports_monotonic_clock()) {
1303     struct timespec tp;
1304     int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1305     assert(status == 0, "gettime error");
1306     jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1307     return result;
1308   } else {
1309     timeval time;
1310     int status = gettimeofday(&time, NULL);
1311     assert(status != -1, "linux error");
1312     jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1313     return 1000 * usecs;
1314   }
1315 }
1316 
1317 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1318   if (os::supports_monotonic_clock()) {
1319     info_ptr->max_value = ALL_64_BITS;
1320 
1321     // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1322     info_ptr->may_skip_backward = false;      // not subject to resetting or drifting
1323     info_ptr->may_skip_forward = false;       // not subject to resetting or drifting
1324   } else {
1325     // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1326     info_ptr->max_value = ALL_64_BITS;
1327 
1328     // gettimeofday is a real time clock so it skips
1329     info_ptr->may_skip_backward = true;
1330     info_ptr->may_skip_forward = true;
1331   }
1332 
1333   info_ptr->kind = JVMTI_TIMER_ELAPSED;                // elapsed not CPU time
1334 }
1335 
1336 // Return the real, user, and system times in seconds from an
1337 // arbitrary fixed point in the past.
1338 bool os::getTimesSecs(double* process_real_time,
1339                       double* process_user_time,
1340                       double* process_system_time) {
1341   struct tms ticks;
1342   clock_t real_ticks = times(&ticks);
1343 
1344   if (real_ticks == (clock_t) (-1)) {
1345     return false;
1346   } else {
1347     double ticks_per_second = (double) clock_tics_per_sec;
1348     *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1349     *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1350     *process_real_time = ((double) real_ticks) / ticks_per_second;
1351 
1352     return true;
1353   }
1354 }
1355 
1356 
1357 char * os::local_time_string(char *buf, size_t buflen) {
1358   struct tm t;
1359   time_t long_time;
1360   time(&long_time);
1361   localtime_r(&long_time, &t);
1362   jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1363                t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1364                t.tm_hour, t.tm_min, t.tm_sec);
1365   return buf;
1366 }
1367 
1368 struct tm* os::localtime_pd(const time_t* clock, struct tm*  res) {
1369   return localtime_r(clock, res);
1370 }
1371 
1372 ////////////////////////////////////////////////////////////////////////////////
1373 // runtime exit support
1374 
1375 // Note: os::shutdown() might be called very early during initialization, or
1376 // called from signal handler. Before adding something to os::shutdown(), make
1377 // sure it is async-safe and can handle partially initialized VM.
1378 void os::shutdown() {
1379 
1380   // allow PerfMemory to attempt cleanup of any persistent resources
1381   perfMemory_exit();
1382 
1383   // needs to remove object in file system
1384   AttachListener::abort();
1385 
1386   // flush buffered output, finish log files
1387   ostream_abort();
1388 
1389   // Check for abort hook
1390   abort_hook_t abort_hook = Arguments::abort_hook();
1391   if (abort_hook != NULL) {
1392     abort_hook();
1393   }
1394 
1395 }
1396 
1397 // Note: os::abort() might be called very early during initialization, or
1398 // called from signal handler. Before adding something to os::abort(), make
1399 // sure it is async-safe and can handle partially initialized VM.
1400 void os::abort(bool dump_core, void* siginfo, const void* context) {
1401   os::shutdown();
1402   if (dump_core) {
1403 #ifndef PRODUCT
1404     fdStream out(defaultStream::output_fd());
1405     out.print_raw("Current thread is ");
1406     char buf[16];
1407     jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1408     out.print_raw_cr(buf);
1409     out.print_raw_cr("Dumping core ...");
1410 #endif
1411     ::abort(); // dump core
1412   }
1413 
1414   ::exit(1);
1415 }
1416 
1417 // Die immediately, no exit hook, no abort hook, no cleanup.
1418 void os::die() {
1419   ::abort();
1420 }
1421 
1422 
1423 // This method is a copy of JDK's sysGetLastErrorString
1424 // from src/solaris/hpi/src/system_md.c
1425 
1426 size_t os::lasterror(char *buf, size_t len) {
1427   if (errno == 0)  return 0;
1428 
1429   const char *s = os::strerror(errno);
1430   size_t n = ::strlen(s);
1431   if (n >= len) {
1432     n = len - 1;
1433   }
1434   ::strncpy(buf, s, n);
1435   buf[n] = '\0';
1436   return n;
1437 }
1438 
1439 // thread_id is kernel thread id (similar to Solaris LWP id)
1440 intx os::current_thread_id() { return os::Linux::gettid(); }
1441 int os::current_process_id() {
1442   return ::getpid();
1443 }
1444 
1445 // DLL functions
1446 
1447 const char* os::dll_file_extension() { return ".so"; }
1448 
1449 // This must be hard coded because it's the system's temporary
1450 // directory not the java application's temp directory, ala java.io.tmpdir.
1451 const char* os::get_temp_directory() { return "/tmp"; }
1452 
1453 static bool file_exists(const char* filename) {
1454   struct stat statbuf;
1455   if (filename == NULL || strlen(filename) == 0) {
1456     return false;
1457   }
1458   return os::stat(filename, &statbuf) == 0;
1459 }
1460 
1461 // check if addr is inside libjvm.so
1462 bool os::address_is_in_vm(address addr) {
1463   static address libjvm_base_addr;
1464   Dl_info dlinfo;
1465 
1466   if (libjvm_base_addr == NULL) {
1467     if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1468       libjvm_base_addr = (address)dlinfo.dli_fbase;
1469     }
1470     assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1471   }
1472 
1473   if (dladdr((void *)addr, &dlinfo) != 0) {
1474     if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1475   }
1476 
1477   return false;
1478 }
1479 
1480 bool os::dll_address_to_function_name(address addr, char *buf,
1481                                       int buflen, int *offset,
1482                                       bool demangle) {
1483   // buf is not optional, but offset is optional
1484   assert(buf != NULL, "sanity check");
1485 
1486   Dl_info dlinfo;
1487 
1488   if (dladdr((void*)addr, &dlinfo) != 0) {
1489     // see if we have a matching symbol
1490     if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1491       if (!(demangle && Decoder::demangle(dlinfo.dli_sname, buf, buflen))) {
1492         jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1493       }
1494       if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1495       return true;
1496     }
1497     // no matching symbol so try for just file info
1498     if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1499       if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1500                           buf, buflen, offset, dlinfo.dli_fname, demangle)) {
1501         return true;
1502       }
1503     }
1504   }
1505 
1506   buf[0] = '\0';
1507   if (offset != NULL) *offset = -1;
1508   return false;
1509 }
1510 
1511 struct _address_to_library_name {
1512   address addr;          // input : memory address
1513   size_t  buflen;        //         size of fname
1514   char*   fname;         // output: library name
1515   address base;          //         library base addr
1516 };
1517 
1518 static int address_to_library_name_callback(struct dl_phdr_info *info,
1519                                             size_t size, void *data) {
1520   int i;
1521   bool found = false;
1522   address libbase = NULL;
1523   struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1524 
1525   // iterate through all loadable segments
1526   for (i = 0; i < info->dlpi_phnum; i++) {
1527     address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1528     if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1529       // base address of a library is the lowest address of its loaded
1530       // segments.
1531       if (libbase == NULL || libbase > segbase) {
1532         libbase = segbase;
1533       }
1534       // see if 'addr' is within current segment
1535       if (segbase <= d->addr &&
1536           d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1537         found = true;
1538       }
1539     }
1540   }
1541 
1542   // dlpi_name is NULL or empty if the ELF file is executable, return 0
1543   // so dll_address_to_library_name() can fall through to use dladdr() which
1544   // can figure out executable name from argv[0].
1545   if (found && info->dlpi_name && info->dlpi_name[0]) {
1546     d->base = libbase;
1547     if (d->fname) {
1548       jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1549     }
1550     return 1;
1551   }
1552   return 0;
1553 }
1554 
1555 bool os::dll_address_to_library_name(address addr, char* buf,
1556                                      int buflen, int* offset) {
1557   // buf is not optional, but offset is optional
1558   assert(buf != NULL, "sanity check");
1559 
1560   Dl_info dlinfo;
1561   struct _address_to_library_name data;
1562 
1563   // There is a bug in old glibc dladdr() implementation that it could resolve
1564   // to wrong library name if the .so file has a base address != NULL. Here
1565   // we iterate through the program headers of all loaded libraries to find
1566   // out which library 'addr' really belongs to. This workaround can be
1567   // removed once the minimum requirement for glibc is moved to 2.3.x.
1568   data.addr = addr;
1569   data.fname = buf;
1570   data.buflen = buflen;
1571   data.base = NULL;
1572   int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1573 
1574   if (rslt) {
1575     // buf already contains library name
1576     if (offset) *offset = addr - data.base;
1577     return true;
1578   }
1579   if (dladdr((void*)addr, &dlinfo) != 0) {
1580     if (dlinfo.dli_fname != NULL) {
1581       jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1582     }
1583     if (dlinfo.dli_fbase != NULL && offset != NULL) {
1584       *offset = addr - (address)dlinfo.dli_fbase;
1585     }
1586     return true;
1587   }
1588 
1589   buf[0] = '\0';
1590   if (offset) *offset = -1;
1591   return false;
1592 }
1593 
1594 // Loads .dll/.so and
1595 // in case of error it checks if .dll/.so was built for the
1596 // same architecture as Hotspot is running on
1597 
1598 
1599 // Remember the stack's state. The Linux dynamic linker will change
1600 // the stack to 'executable' at most once, so we must safepoint only once.
1601 bool os::Linux::_stack_is_executable = false;
1602 
1603 // VM operation that loads a library.  This is necessary if stack protection
1604 // of the Java stacks can be lost during loading the library.  If we
1605 // do not stop the Java threads, they can stack overflow before the stacks
1606 // are protected again.
1607 class VM_LinuxDllLoad: public VM_Operation {
1608  private:
1609   const char *_filename;
1610   char *_ebuf;
1611   int _ebuflen;
1612   void *_lib;
1613  public:
1614   VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1615     _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1616   VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1617   void doit() {
1618     _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1619     os::Linux::_stack_is_executable = true;
1620   }
1621   void* loaded_library() { return _lib; }
1622 };
1623 
1624 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1625   void * result = NULL;
1626   bool load_attempted = false;
1627 
1628   // Check whether the library to load might change execution rights
1629   // of the stack. If they are changed, the protection of the stack
1630   // guard pages will be lost. We need a safepoint to fix this.
1631   //
1632   // See Linux man page execstack(8) for more info.
1633   if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1634     if (!ElfFile::specifies_noexecstack(filename)) {
1635       if (!is_init_completed()) {
1636         os::Linux::_stack_is_executable = true;
1637         // This is OK - No Java threads have been created yet, and hence no
1638         // stack guard pages to fix.
1639         //
1640         // This should happen only when you are building JDK7 using a very
1641         // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1642         //
1643         // Dynamic loader will make all stacks executable after
1644         // this function returns, and will not do that again.
1645         assert(Threads::first() == NULL, "no Java threads should exist yet.");
1646       } else {
1647         warning("You have loaded library %s which might have disabled stack guard. "
1648                 "The VM will try to fix the stack guard now.\n"
1649                 "It's highly recommended that you fix the library with "
1650                 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1651                 filename);
1652 
1653         assert(Thread::current()->is_Java_thread(), "must be Java thread");
1654         JavaThread *jt = JavaThread::current();
1655         if (jt->thread_state() != _thread_in_native) {
1656           // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1657           // that requires ExecStack. Cannot enter safe point. Let's give up.
1658           warning("Unable to fix stack guard. Giving up.");
1659         } else {
1660           if (!LoadExecStackDllInVMThread) {
1661             // This is for the case where the DLL has an static
1662             // constructor function that executes JNI code. We cannot
1663             // load such DLLs in the VMThread.
1664             result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1665           }
1666 
1667           ThreadInVMfromNative tiv(jt);
1668           debug_only(VMNativeEntryWrapper vew;)
1669 
1670           VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1671           VMThread::execute(&op);
1672           if (LoadExecStackDllInVMThread) {
1673             result = op.loaded_library();
1674           }
1675           load_attempted = true;
1676         }
1677       }
1678     }
1679   }
1680 
1681   if (!load_attempted) {
1682     result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1683   }
1684 
1685   if (result != NULL) {
1686     // Successful loading
1687     return result;
1688   }
1689 
1690   Elf32_Ehdr elf_head;
1691   int diag_msg_max_length=ebuflen-strlen(ebuf);
1692   char* diag_msg_buf=ebuf+strlen(ebuf);
1693 
1694   if (diag_msg_max_length==0) {
1695     // No more space in ebuf for additional diagnostics message
1696     return NULL;
1697   }
1698 
1699 
1700   int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1701 
1702   if (file_descriptor < 0) {
1703     // Can't open library, report dlerror() message
1704     return NULL;
1705   }
1706 
1707   bool failed_to_read_elf_head=
1708     (sizeof(elf_head)!=
1709      (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1710 
1711   ::close(file_descriptor);
1712   if (failed_to_read_elf_head) {
1713     // file i/o error - report dlerror() msg
1714     return NULL;
1715   }
1716 
1717   typedef struct {
1718     Elf32_Half    code;         // Actual value as defined in elf.h
1719     Elf32_Half    compat_class; // Compatibility of archs at VM's sense
1720     unsigned char elf_class;    // 32 or 64 bit
1721     unsigned char endianess;    // MSB or LSB
1722     char*         name;         // String representation
1723   } arch_t;
1724 
1725 #ifndef EM_486
1726   #define EM_486          6               /* Intel 80486 */
1727 #endif
1728 #ifndef EM_AARCH64
1729   #define EM_AARCH64    183               /* ARM AARCH64 */
1730 #endif
1731 
1732   static const arch_t arch_array[]={
1733     {EM_386,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1734     {EM_486,         EM_386,     ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1735     {EM_IA_64,       EM_IA_64,   ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1736     {EM_X86_64,      EM_X86_64,  ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1737     {EM_SPARC,       EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1738     {EM_SPARC32PLUS, EM_SPARC,   ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1739     {EM_SPARCV9,     EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1740     {EM_PPC,         EM_PPC,     ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1741 #if defined(VM_LITTLE_ENDIAN)
1742     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1743     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2LSB, (char*)"SuperH"},
1744 #else
1745     {EM_PPC64,       EM_PPC64,   ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1746     {EM_SH,          EM_SH,      ELFCLASS32, ELFDATA2MSB, (char*)"SuperH BE"},
1747 #endif
1748     {EM_ARM,         EM_ARM,     ELFCLASS32,   ELFDATA2LSB, (char*)"ARM"},
1749     {EM_S390,        EM_S390,    ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1750     {EM_ALPHA,       EM_ALPHA,   ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1751     {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1752     {EM_MIPS,        EM_MIPS,    ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1753     {EM_PARISC,      EM_PARISC,  ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1754     {EM_68K,         EM_68K,     ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1755     {EM_AARCH64,     EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1756   };
1757 
1758 #if  (defined IA32)
1759   static  Elf32_Half running_arch_code=EM_386;
1760 #elif   (defined AMD64)
1761   static  Elf32_Half running_arch_code=EM_X86_64;
1762 #elif  (defined IA64)
1763   static  Elf32_Half running_arch_code=EM_IA_64;
1764 #elif  (defined __sparc) && (defined _LP64)
1765   static  Elf32_Half running_arch_code=EM_SPARCV9;
1766 #elif  (defined __sparc) && (!defined _LP64)
1767   static  Elf32_Half running_arch_code=EM_SPARC;
1768 #elif  (defined __powerpc64__)
1769   static  Elf32_Half running_arch_code=EM_PPC64;
1770 #elif  (defined __powerpc__)
1771   static  Elf32_Half running_arch_code=EM_PPC;
1772 #elif  (defined AARCH64)
1773   static  Elf32_Half running_arch_code=EM_AARCH64;
1774 #elif  (defined ARM)
1775   static  Elf32_Half running_arch_code=EM_ARM;
1776 #elif  (defined S390)
1777   static  Elf32_Half running_arch_code=EM_S390;
1778 #elif  (defined ALPHA)
1779   static  Elf32_Half running_arch_code=EM_ALPHA;
1780 #elif  (defined MIPSEL)
1781   static  Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1782 #elif  (defined PARISC)
1783   static  Elf32_Half running_arch_code=EM_PARISC;
1784 #elif  (defined MIPS)
1785   static  Elf32_Half running_arch_code=EM_MIPS;
1786 #elif  (defined M68K)
1787   static  Elf32_Half running_arch_code=EM_68K;
1788 #elif  (defined SH)
1789   static  Elf32_Half running_arch_code=EM_SH;
1790 #else
1791     #error Method os::dll_load requires that one of following is defined:\
1792         AARCH64, ALPHA, ARM, AMD64, IA32, IA64, M68K, MIPS, MIPSEL, PARISC, __powerpc__, __powerpc64__, S390, SH, __sparc
1793 #endif
1794 
1795   // Identify compatability class for VM's architecture and library's architecture
1796   // Obtain string descriptions for architectures
1797 
1798   arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1799   int running_arch_index=-1;
1800 
1801   for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1802     if (running_arch_code == arch_array[i].code) {
1803       running_arch_index    = i;
1804     }
1805     if (lib_arch.code == arch_array[i].code) {
1806       lib_arch.compat_class = arch_array[i].compat_class;
1807       lib_arch.name         = arch_array[i].name;
1808     }
1809   }
1810 
1811   assert(running_arch_index != -1,
1812          "Didn't find running architecture code (running_arch_code) in arch_array");
1813   if (running_arch_index == -1) {
1814     // Even though running architecture detection failed
1815     // we may still continue with reporting dlerror() message
1816     return NULL;
1817   }
1818 
1819   if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1820     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1821     return NULL;
1822   }
1823 
1824 #ifndef S390
1825   if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1826     ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1827     return NULL;
1828   }
1829 #endif // !S390
1830 
1831   if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1832     if (lib_arch.name!=NULL) {
1833       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1834                  " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1835                  lib_arch.name, arch_array[running_arch_index].name);
1836     } else {
1837       ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1838                  " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1839                  lib_arch.code,
1840                  arch_array[running_arch_index].name);
1841     }
1842   }
1843 
1844   return NULL;
1845 }
1846 
1847 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1848                                 int ebuflen) {
1849   void * result = ::dlopen(filename, RTLD_LAZY);
1850   if (result == NULL) {
1851     ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1852     ebuf[ebuflen-1] = '\0';
1853   }
1854   return result;
1855 }
1856 
1857 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1858                                        int ebuflen) {
1859   void * result = NULL;
1860   if (LoadExecStackDllInVMThread) {
1861     result = dlopen_helper(filename, ebuf, ebuflen);
1862   }
1863 
1864   // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
1865   // library that requires an executable stack, or which does not have this
1866   // stack attribute set, dlopen changes the stack attribute to executable. The
1867   // read protection of the guard pages gets lost.
1868   //
1869   // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
1870   // may have been queued at the same time.
1871 
1872   if (!_stack_is_executable) {
1873     JavaThread *jt = Threads::first();
1874 
1875     while (jt) {
1876       if (!jt->stack_guard_zone_unused() &&     // Stack not yet fully initialized
1877           jt->stack_guards_enabled()) {         // No pending stack overflow exceptions
1878         if (!os::guard_memory((char *)jt->stack_end(), jt->stack_guard_zone_size())) {
1879           warning("Attempt to reguard stack yellow zone failed.");
1880         }
1881       }
1882       jt = jt->next();
1883     }
1884   }
1885 
1886   return result;
1887 }
1888 
1889 void* os::dll_lookup(void* handle, const char* name) {
1890   void* res = dlsym(handle, name);
1891   return res;
1892 }
1893 
1894 void* os::get_default_process_handle() {
1895   return (void*)::dlopen(NULL, RTLD_LAZY);
1896 }
1897 
1898 static bool _print_ascii_file(const char* filename, outputStream* st) {
1899   int fd = ::open(filename, O_RDONLY);
1900   if (fd == -1) {
1901     return false;
1902   }
1903 
1904   char buf[33];
1905   int bytes;
1906   buf[32] = '\0';
1907   while ((bytes = ::read(fd, buf, sizeof(buf)-1)) > 0) {
1908     st->print_raw(buf, bytes);
1909   }
1910 
1911   ::close(fd);
1912 
1913   return true;
1914 }
1915 
1916 void os::print_dll_info(outputStream *st) {
1917   st->print_cr("Dynamic libraries:");
1918 
1919   char fname[32];
1920   pid_t pid = os::Linux::gettid();
1921 
1922   jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1923 
1924   if (!_print_ascii_file(fname, st)) {
1925     st->print("Can not get library information for pid = %d\n", pid);
1926   }
1927 }
1928 
1929 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
1930   FILE *procmapsFile = NULL;
1931 
1932   // Open the procfs maps file for the current process
1933   if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
1934     // Allocate PATH_MAX for file name plus a reasonable size for other fields.
1935     char line[PATH_MAX + 100];
1936 
1937     // Read line by line from 'file'
1938     while (fgets(line, sizeof(line), procmapsFile) != NULL) {
1939       u8 base, top, offset, inode;
1940       char permissions[5];
1941       char device[6];
1942       char name[PATH_MAX + 1];
1943 
1944       // Parse fields from line
1945       sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %5s " INT64_FORMAT " %s",
1946              &base, &top, permissions, &offset, device, &inode, name);
1947 
1948       // Filter by device id '00:00' so that we only get file system mapped files.
1949       if (strcmp(device, "00:00") != 0) {
1950 
1951         // Call callback with the fields of interest
1952         if(callback(name, (address)base, (address)top, param)) {
1953           // Oops abort, callback aborted
1954           fclose(procmapsFile);
1955           return 1;
1956         }
1957       }
1958     }
1959     fclose(procmapsFile);
1960   }
1961   return 0;
1962 }
1963 
1964 void os::print_os_info_brief(outputStream* st) {
1965   os::Linux::print_distro_info(st);
1966 
1967   os::Posix::print_uname_info(st);
1968 
1969   os::Linux::print_libversion_info(st);
1970 
1971 }
1972 
1973 void os::print_os_info(outputStream* st) {
1974   st->print("OS:");
1975 
1976   os::Linux::print_distro_info(st);
1977 
1978   os::Posix::print_uname_info(st);
1979 
1980   // Print warning if unsafe chroot environment detected
1981   if (unsafe_chroot_detected) {
1982     st->print("WARNING!! ");
1983     st->print_cr("%s", unstable_chroot_error);
1984   }
1985 
1986   os::Linux::print_libversion_info(st);
1987 
1988   os::Posix::print_rlimit_info(st);
1989 
1990   os::Posix::print_load_average(st);
1991 
1992   os::Linux::print_full_memory_info(st);
1993 
1994   os::Linux::print_container_info(st);
1995 }
1996 
1997 // Try to identify popular distros.
1998 // Most Linux distributions have a /etc/XXX-release file, which contains
1999 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2000 // file that also contains the OS version string. Some have more than one
2001 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2002 // /etc/redhat-release.), so the order is important.
2003 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2004 // their own specific XXX-release file as well as a redhat-release file.
2005 // Because of this the XXX-release file needs to be searched for before the
2006 // redhat-release file.
2007 // Since Red Hat and SuSE have an lsb-release file that is not very descriptive the
2008 // search for redhat-release / SuSE-release needs to be before lsb-release.
2009 // Since the lsb-release file is the new standard it needs to be searched
2010 // before the older style release files.
2011 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2012 // next to last resort.  The os-release file is a new standard that contains
2013 // distribution information and the system-release file seems to be an old
2014 // standard that has been replaced by the lsb-release and os-release files.
2015 // Searching for the debian_version file is the last resort.  It contains
2016 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2017 // "Debian " is printed before the contents of the debian_version file.
2018 
2019 const char* distro_files[] = {
2020   "/etc/oracle-release",
2021   "/etc/mandriva-release",
2022   "/etc/mandrake-release",
2023   "/etc/sun-release",
2024   "/etc/redhat-release",
2025   "/etc/SuSE-release",
2026   "/etc/lsb-release",
2027   "/etc/turbolinux-release",
2028   "/etc/gentoo-release",
2029   "/etc/ltib-release",
2030   "/etc/angstrom-version",
2031   "/etc/system-release",
2032   "/etc/os-release",
2033   NULL };
2034 
2035 void os::Linux::print_distro_info(outputStream* st) {
2036   for (int i = 0;; i++) {
2037     const char* file = distro_files[i];
2038     if (file == NULL) {
2039       break;  // done
2040     }
2041     // If file prints, we found it.
2042     if (_print_ascii_file(file, st)) {
2043       return;
2044     }
2045   }
2046 
2047   if (file_exists("/etc/debian_version")) {
2048     st->print("Debian ");
2049     _print_ascii_file("/etc/debian_version", st);
2050   } else {
2051     st->print("Linux");
2052   }
2053   st->cr();
2054 }
2055 
2056 static void parse_os_info_helper(FILE* fp, char* distro, size_t length, bool get_first_line) {
2057   char buf[256];
2058   while (fgets(buf, sizeof(buf), fp)) {
2059     // Edit out extra stuff in expected format
2060     if (strstr(buf, "DISTRIB_DESCRIPTION=") != NULL || strstr(buf, "PRETTY_NAME=") != NULL) {
2061       char* ptr = strstr(buf, "\"");  // the name is in quotes
2062       if (ptr != NULL) {
2063         ptr++; // go beyond first quote
2064         char* nl = strchr(ptr, '\"');
2065         if (nl != NULL) *nl = '\0';
2066         strncpy(distro, ptr, length);
2067       } else {
2068         ptr = strstr(buf, "=");
2069         ptr++; // go beyond equals then
2070         char* nl = strchr(ptr, '\n');
2071         if (nl != NULL) *nl = '\0';
2072         strncpy(distro, ptr, length);
2073       }
2074       return;
2075     } else if (get_first_line) {
2076       char* nl = strchr(buf, '\n');
2077       if (nl != NULL) *nl = '\0';
2078       strncpy(distro, buf, length);
2079       return;
2080     }
2081   }
2082   // print last line and close
2083   char* nl = strchr(buf, '\n');
2084   if (nl != NULL) *nl = '\0';
2085   strncpy(distro, buf, length);
2086 }
2087 
2088 static void parse_os_info(char* distro, size_t length, const char* file) {
2089   FILE* fp = fopen(file, "r");
2090   if (fp != NULL) {
2091     // if suse format, print out first line
2092     bool get_first_line = (strcmp(file, "/etc/SuSE-release") == 0);
2093     parse_os_info_helper(fp, distro, length, get_first_line);
2094     fclose(fp);
2095   }
2096 }
2097 
2098 void os::get_summary_os_info(char* buf, size_t buflen) {
2099   for (int i = 0;; i++) {
2100     const char* file = distro_files[i];
2101     if (file == NULL) {
2102       break; // ran out of distro_files
2103     }
2104     if (file_exists(file)) {
2105       parse_os_info(buf, buflen, file);
2106       return;
2107     }
2108   }
2109   // special case for debian
2110   if (file_exists("/etc/debian_version")) {
2111     strncpy(buf, "Debian ", buflen);
2112     parse_os_info(&buf[7], buflen-7, "/etc/debian_version");
2113   } else {
2114     strncpy(buf, "Linux", buflen);
2115   }
2116 }
2117 
2118 void os::Linux::print_libversion_info(outputStream* st) {
2119   // libc, pthread
2120   st->print("libc:");
2121   st->print("%s ", os::Linux::glibc_version());
2122   st->print("%s ", os::Linux::libpthread_version());
2123   st->cr();
2124 }
2125 
2126 void os::Linux::print_full_memory_info(outputStream* st) {
2127   st->print("\n/proc/meminfo:\n");
2128   _print_ascii_file("/proc/meminfo", st);
2129   st->cr();
2130 }
2131 
2132 void os::Linux::print_container_info(outputStream* st) {
2133   if (OSContainer::is_containerized()) {
2134     st->print("container (cgroup) information:\n");
2135 
2136     char *p = OSContainer::container_type();
2137     if (p == NULL)
2138       st->print("container_type() failed\n");
2139     else {
2140       st->print("container_type: %s\n", p);
2141     }
2142 
2143     p = OSContainer::cpu_cpuset_cpus();
2144     if (p == NULL)
2145       st->print("cpu_cpuset_cpus() failed\n");
2146     else {
2147       st->print("cpu_cpuset_cpus: %s\n", p);
2148       free(p);
2149     }
2150 
2151     p = OSContainer::cpu_cpuset_memory_nodes();
2152     if (p < 0)
2153       st->print("cpu_memory_nodes() failed\n");
2154     else {
2155       st->print("cpu_memory_nodes: %s\n", p);
2156       free(p);
2157     }
2158 
2159     int i = OSContainer::active_processor_count();
2160     if (i < 0)
2161       st->print("active_processor_count() failed\n");
2162     else
2163       st->print("active_processor_count: %d\n", i);
2164 
2165     i = OSContainer::cpu_quota();
2166     st->print("cpu_quota: %d\n", i);
2167 
2168     i = OSContainer::cpu_period();
2169     st->print("cpu_period: %d\n", i);
2170 
2171     i = OSContainer::cpu_shares();
2172     st->print("cpu_shares: %d\n", i);
2173 
2174     jlong j = OSContainer::memory_limit_in_bytes();
2175     st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2176 
2177     j = OSContainer::memory_and_swap_limit_in_bytes();
2178     st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2179 
2180     j = OSContainer::memory_soft_limit_in_bytes();
2181     st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2182 
2183     j = OSContainer::OSContainer::memory_usage_in_bytes();
2184     st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2185 
2186     j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2187     st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2188     st->cr();
2189   }
2190 }
2191 
2192 void os::print_memory_info(outputStream* st) {
2193 
2194   st->print("Memory:");
2195   st->print(" %dk page", os::vm_page_size()>>10);
2196 
2197   // values in struct sysinfo are "unsigned long"
2198   struct sysinfo si;
2199   sysinfo(&si);
2200 
2201   st->print(", physical " UINT64_FORMAT "k",
2202             os::physical_memory() >> 10);
2203   st->print("(" UINT64_FORMAT "k free)",
2204             os::available_memory() >> 10);
2205   st->print(", swap " UINT64_FORMAT "k",
2206             ((jlong)si.totalswap * si.mem_unit) >> 10);
2207   st->print("(" UINT64_FORMAT "k free)",
2208             ((jlong)si.freeswap * si.mem_unit) >> 10);
2209   st->cr();
2210 }
2211 
2212 // Print the first "model name" line and the first "flags" line
2213 // that we find and nothing more. We assume "model name" comes
2214 // before "flags" so if we find a second "model name", then the
2215 // "flags" field is considered missing.
2216 static bool print_model_name_and_flags(outputStream* st, char* buf, size_t buflen) {
2217 #if defined(IA32) || defined(AMD64)
2218   // Other platforms have less repetitive cpuinfo files
2219   FILE *fp = fopen("/proc/cpuinfo", "r");
2220   if (fp) {
2221     while (!feof(fp)) {
2222       if (fgets(buf, buflen, fp)) {
2223         // Assume model name comes before flags
2224         bool model_name_printed = false;
2225         if (strstr(buf, "model name") != NULL) {
2226           if (!model_name_printed) {
2227             st->print_raw("CPU Model and flags from /proc/cpuinfo:\n");
2228             st->print_raw(buf);
2229             model_name_printed = true;
2230           } else {
2231             // model name printed but not flags?  Odd, just return
2232             fclose(fp);
2233             return true;
2234           }
2235         }
2236         // print the flags line too
2237         if (strstr(buf, "flags") != NULL) {
2238           st->print_raw(buf);
2239           fclose(fp);
2240           return true;
2241         }
2242       }
2243     }
2244     fclose(fp);
2245   }
2246 #endif // x86 platforms
2247   return false;
2248 }
2249 
2250 void os::pd_print_cpu_info(outputStream* st, char* buf, size_t buflen) {
2251   // Only print the model name if the platform provides this as a summary
2252   if (!print_model_name_and_flags(st, buf, buflen)) {
2253     st->print("\n/proc/cpuinfo:\n");
2254     if (!_print_ascii_file("/proc/cpuinfo", st)) {
2255       st->print_cr("  <Not Available>");
2256     }
2257   }
2258 }
2259 
2260 #if defined(AMD64) || defined(IA32) || defined(X32)
2261 const char* search_string = "model name";
2262 #elif defined(M68K)
2263 const char* search_string = "CPU";
2264 #elif defined(PPC64)
2265 const char* search_string = "cpu";
2266 #elif defined(S390)
2267 const char* search_string = "processor";
2268 #elif defined(SPARC)
2269 const char* search_string = "cpu";
2270 #else
2271 const char* search_string = "Processor";
2272 #endif
2273 
2274 // Parses the cpuinfo file for string representing the model name.
2275 void os::get_summary_cpu_info(char* cpuinfo, size_t length) {
2276   FILE* fp = fopen("/proc/cpuinfo", "r");
2277   if (fp != NULL) {
2278     while (!feof(fp)) {
2279       char buf[256];
2280       if (fgets(buf, sizeof(buf), fp)) {
2281         char* start = strstr(buf, search_string);
2282         if (start != NULL) {
2283           char *ptr = start + strlen(search_string);
2284           char *end = buf + strlen(buf);
2285           while (ptr != end) {
2286              // skip whitespace and colon for the rest of the name.
2287              if (*ptr != ' ' && *ptr != '\t' && *ptr != ':') {
2288                break;
2289              }
2290              ptr++;
2291           }
2292           if (ptr != end) {
2293             // reasonable string, get rid of newline and keep the rest
2294             char* nl = strchr(buf, '\n');
2295             if (nl != NULL) *nl = '\0';
2296             strncpy(cpuinfo, ptr, length);
2297             fclose(fp);
2298             return;
2299           }
2300         }
2301       }
2302     }
2303     fclose(fp);
2304   }
2305   // cpuinfo not found or parsing failed, just print generic string.  The entire
2306   // /proc/cpuinfo file will be printed later in the file (or enough of it for x86)
2307 #if   defined(AARCH64)
2308   strncpy(cpuinfo, "AArch64", length);
2309 #elif defined(AMD64)
2310   strncpy(cpuinfo, "x86_64", length);
2311 #elif defined(ARM)  // Order wrt. AARCH64 is relevant!
2312   strncpy(cpuinfo, "ARM", length);
2313 #elif defined(IA32)
2314   strncpy(cpuinfo, "x86_32", length);
2315 #elif defined(IA64)
2316   strncpy(cpuinfo, "IA64", length);
2317 #elif defined(PPC)
2318   strncpy(cpuinfo, "PPC64", length);
2319 #elif defined(S390)
2320   strncpy(cpuinfo, "S390", length);
2321 #elif defined(SPARC)
2322   strncpy(cpuinfo, "sparcv9", length);
2323 #elif defined(ZERO_LIBARCH)
2324   strncpy(cpuinfo, ZERO_LIBARCH, length);
2325 #else
2326   strncpy(cpuinfo, "unknown", length);
2327 #endif
2328 }
2329 
2330 static void print_signal_handler(outputStream* st, int sig,
2331                                  char* buf, size_t buflen);
2332 
2333 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2334   st->print_cr("Signal Handlers:");
2335   print_signal_handler(st, SIGSEGV, buf, buflen);
2336   print_signal_handler(st, SIGBUS , buf, buflen);
2337   print_signal_handler(st, SIGFPE , buf, buflen);
2338   print_signal_handler(st, SIGPIPE, buf, buflen);
2339   print_signal_handler(st, SIGXFSZ, buf, buflen);
2340   print_signal_handler(st, SIGILL , buf, buflen);
2341   print_signal_handler(st, SR_signum, buf, buflen);
2342   print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2343   print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2344   print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2345   print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2346 #if defined(PPC64)
2347   print_signal_handler(st, SIGTRAP, buf, buflen);
2348 #endif
2349 }
2350 
2351 static char saved_jvm_path[MAXPATHLEN] = {0};
2352 
2353 // Find the full path to the current module, libjvm.so
2354 void os::jvm_path(char *buf, jint buflen) {
2355   // Error checking.
2356   if (buflen < MAXPATHLEN) {
2357     assert(false, "must use a large-enough buffer");
2358     buf[0] = '\0';
2359     return;
2360   }
2361   // Lazy resolve the path to current module.
2362   if (saved_jvm_path[0] != 0) {
2363     strcpy(buf, saved_jvm_path);
2364     return;
2365   }
2366 
2367   char dli_fname[MAXPATHLEN];
2368   bool ret = dll_address_to_library_name(
2369                                          CAST_FROM_FN_PTR(address, os::jvm_path),
2370                                          dli_fname, sizeof(dli_fname), NULL);
2371   assert(ret, "cannot locate libjvm");
2372   char *rp = NULL;
2373   if (ret && dli_fname[0] != '\0') {
2374     rp = os::Posix::realpath(dli_fname, buf, buflen);
2375   }
2376   if (rp == NULL) {
2377     return;
2378   }
2379 
2380   if (Arguments::sun_java_launcher_is_altjvm()) {
2381     // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2382     // value for buf is "<JAVA_HOME>/jre/lib/<vmtype>/libjvm.so".
2383     // If "/jre/lib/" appears at the right place in the string, then
2384     // assume we are installed in a JDK and we're done. Otherwise, check
2385     // for a JAVA_HOME environment variable and fix up the path so it
2386     // looks like libjvm.so is installed there (append a fake suffix
2387     // hotspot/libjvm.so).
2388     const char *p = buf + strlen(buf) - 1;
2389     for (int count = 0; p > buf && count < 5; ++count) {
2390       for (--p; p > buf && *p != '/'; --p)
2391         /* empty */ ;
2392     }
2393 
2394     if (strncmp(p, "/jre/lib/", 9) != 0) {
2395       // Look for JAVA_HOME in the environment.
2396       char* java_home_var = ::getenv("JAVA_HOME");
2397       if (java_home_var != NULL && java_home_var[0] != 0) {
2398         char* jrelib_p;
2399         int len;
2400 
2401         // Check the current module name "libjvm.so".
2402         p = strrchr(buf, '/');
2403         if (p == NULL) {
2404           return;
2405         }
2406         assert(strstr(p, "/libjvm") == p, "invalid library name");
2407 
2408         rp = os::Posix::realpath(java_home_var, buf, buflen);
2409         if (rp == NULL) {
2410           return;
2411         }
2412 
2413         // determine if this is a legacy image or modules image
2414         // modules image doesn't have "jre" subdirectory
2415         len = strlen(buf);
2416         assert(len < buflen, "Ran out of buffer room");
2417         jrelib_p = buf + len;
2418         snprintf(jrelib_p, buflen-len, "/jre/lib");
2419         if (0 != access(buf, F_OK)) {
2420           snprintf(jrelib_p, buflen-len, "/lib");
2421         }
2422 
2423         if (0 == access(buf, F_OK)) {
2424           // Use current module name "libjvm.so"
2425           len = strlen(buf);
2426           snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2427         } else {
2428           // Go back to path of .so
2429           rp = os::Posix::realpath(dli_fname, buf, buflen);
2430           if (rp == NULL) {
2431             return;
2432           }
2433         }
2434       }
2435     }
2436   }
2437 
2438   strncpy(saved_jvm_path, buf, MAXPATHLEN);
2439   saved_jvm_path[MAXPATHLEN - 1] = '\0';
2440 }
2441 
2442 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2443   // no prefix required, not even "_"
2444 }
2445 
2446 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2447   // no suffix required
2448 }
2449 
2450 ////////////////////////////////////////////////////////////////////////////////
2451 // sun.misc.Signal support
2452 
2453 static volatile jint sigint_count = 0;
2454 
2455 static void UserHandler(int sig, void *siginfo, void *context) {
2456   // 4511530 - sem_post is serialized and handled by the manager thread. When
2457   // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2458   // don't want to flood the manager thread with sem_post requests.
2459   if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2460     return;
2461   }
2462 
2463   // Ctrl-C is pressed during error reporting, likely because the error
2464   // handler fails to abort. Let VM die immediately.
2465   if (sig == SIGINT && VMError::is_error_reported()) {
2466     os::die();
2467   }
2468 
2469   os::signal_notify(sig);
2470 }
2471 
2472 void* os::user_handler() {
2473   return CAST_FROM_FN_PTR(void*, UserHandler);
2474 }
2475 
2476 struct timespec PosixSemaphore::create_timespec(unsigned int sec, int nsec) {
2477   struct timespec ts;
2478   // Semaphore's are always associated with CLOCK_REALTIME
2479   os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2480   // see unpackTime for discussion on overflow checking
2481   if (sec >= MAX_SECS) {
2482     ts.tv_sec += MAX_SECS;
2483     ts.tv_nsec = 0;
2484   } else {
2485     ts.tv_sec += sec;
2486     ts.tv_nsec += nsec;
2487     if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2488       ts.tv_nsec -= NANOSECS_PER_SEC;
2489       ++ts.tv_sec; // note: this must be <= max_secs
2490     }
2491   }
2492 
2493   return ts;
2494 }
2495 
2496 extern "C" {
2497   typedef void (*sa_handler_t)(int);
2498   typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2499 }
2500 
2501 void* os::signal(int signal_number, void* handler) {
2502   struct sigaction sigAct, oldSigAct;
2503 
2504   sigfillset(&(sigAct.sa_mask));
2505   sigAct.sa_flags   = SA_RESTART|SA_SIGINFO;
2506   sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2507 
2508   if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2509     // -1 means registration failed
2510     return (void *)-1;
2511   }
2512 
2513   return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2514 }
2515 
2516 void os::signal_raise(int signal_number) {
2517   ::raise(signal_number);
2518 }
2519 
2520 // The following code is moved from os.cpp for making this
2521 // code platform specific, which it is by its very nature.
2522 
2523 // Will be modified when max signal is changed to be dynamic
2524 int os::sigexitnum_pd() {
2525   return NSIG;
2526 }
2527 
2528 // a counter for each possible signal value
2529 static volatile jint pending_signals[NSIG+1] = { 0 };
2530 
2531 // Linux(POSIX) specific hand shaking semaphore.
2532 static sem_t sig_sem;
2533 static PosixSemaphore sr_semaphore;
2534 
2535 void os::signal_init_pd() {
2536   // Initialize signal structures
2537   ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2538 
2539   // Initialize signal semaphore
2540   ::sem_init(&sig_sem, 0, 0);
2541 }
2542 
2543 void os::signal_notify(int sig) {
2544   Atomic::inc(&pending_signals[sig]);
2545   ::sem_post(&sig_sem);
2546 }
2547 
2548 static int check_pending_signals(bool wait) {
2549   Atomic::store(0, &sigint_count);
2550   for (;;) {
2551     for (int i = 0; i < NSIG + 1; i++) {
2552       jint n = pending_signals[i];
2553       if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2554         return i;
2555       }
2556     }
2557     if (!wait) {
2558       return -1;
2559     }
2560     JavaThread *thread = JavaThread::current();
2561     ThreadBlockInVM tbivm(thread);
2562 
2563     bool threadIsSuspended;
2564     do {
2565       thread->set_suspend_equivalent();
2566       // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2567       ::sem_wait(&sig_sem);
2568 
2569       // were we externally suspended while we were waiting?
2570       threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2571       if (threadIsSuspended) {
2572         // The semaphore has been incremented, but while we were waiting
2573         // another thread suspended us. We don't want to continue running
2574         // while suspended because that would surprise the thread that
2575         // suspended us.
2576         ::sem_post(&sig_sem);
2577 
2578         thread->java_suspend_self();
2579       }
2580     } while (threadIsSuspended);
2581   }
2582 }
2583 
2584 int os::signal_lookup() {
2585   return check_pending_signals(false);
2586 }
2587 
2588 int os::signal_wait() {
2589   return check_pending_signals(true);
2590 }
2591 
2592 ////////////////////////////////////////////////////////////////////////////////
2593 // Virtual Memory
2594 
2595 int os::vm_page_size() {
2596   // Seems redundant as all get out
2597   assert(os::Linux::page_size() != -1, "must call os::init");
2598   return os::Linux::page_size();
2599 }
2600 
2601 // Solaris allocates memory by pages.
2602 int os::vm_allocation_granularity() {
2603   assert(os::Linux::page_size() != -1, "must call os::init");
2604   return os::Linux::page_size();
2605 }
2606 
2607 // Rationale behind this function:
2608 //  current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2609 //  mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2610 //  samples for JITted code. Here we create private executable mapping over the code cache
2611 //  and then we can use standard (well, almost, as mapping can change) way to provide
2612 //  info for the reporting script by storing timestamp and location of symbol
2613 void linux_wrap_code(char* base, size_t size) {
2614   static volatile jint cnt = 0;
2615 
2616   if (!UseOprofile) {
2617     return;
2618   }
2619 
2620   char buf[PATH_MAX+1];
2621   int num = Atomic::add(1, &cnt);
2622 
2623   snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2624            os::get_temp_directory(), os::current_process_id(), num);
2625   unlink(buf);
2626 
2627   int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2628 
2629   if (fd != -1) {
2630     off_t rv = ::lseek(fd, size-2, SEEK_SET);
2631     if (rv != (off_t)-1) {
2632       if (::write(fd, "", 1) == 1) {
2633         mmap(base, size,
2634              PROT_READ|PROT_WRITE|PROT_EXEC,
2635              MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2636       }
2637     }
2638     ::close(fd);
2639     unlink(buf);
2640   }
2641 }
2642 
2643 static bool recoverable_mmap_error(int err) {
2644   // See if the error is one we can let the caller handle. This
2645   // list of errno values comes from JBS-6843484. I can't find a
2646   // Linux man page that documents this specific set of errno
2647   // values so while this list currently matches Solaris, it may
2648   // change as we gain experience with this failure mode.
2649   switch (err) {
2650   case EBADF:
2651   case EINVAL:
2652   case ENOTSUP:
2653     // let the caller deal with these errors
2654     return true;
2655 
2656   default:
2657     // Any remaining errors on this OS can cause our reserved mapping
2658     // to be lost. That can cause confusion where different data
2659     // structures think they have the same memory mapped. The worst
2660     // scenario is if both the VM and a library think they have the
2661     // same memory mapped.
2662     return false;
2663   }
2664 }
2665 
2666 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2667                                     int err) {
2668   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2669           ", %d) failed; error='%s' (errno=%d)", p2i(addr), size, exec,
2670           os::strerror(err), err);
2671 }
2672 
2673 static void warn_fail_commit_memory(char* addr, size_t size,
2674                                     size_t alignment_hint, bool exec,
2675                                     int err) {
2676   warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2677           ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", p2i(addr), size,
2678           alignment_hint, exec, os::strerror(err), err);
2679 }
2680 
2681 // NOTE: Linux kernel does not really reserve the pages for us.
2682 //       All it does is to check if there are enough free pages
2683 //       left at the time of mmap(). This could be a potential
2684 //       problem.
2685 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2686   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2687   uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2688                                      MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2689   if (res != (uintptr_t) MAP_FAILED) {
2690     if (UseNUMAInterleaving) {
2691       numa_make_global(addr, size);
2692     }
2693     return 0;
2694   }
2695 
2696   int err = errno;  // save errno from mmap() call above
2697 
2698   if (!recoverable_mmap_error(err)) {
2699     warn_fail_commit_memory(addr, size, exec, err);
2700     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2701   }
2702 
2703   return err;
2704 }
2705 
2706 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2707   return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2708 }
2709 
2710 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2711                                   const char* mesg) {
2712   assert(mesg != NULL, "mesg must be specified");
2713   int err = os::Linux::commit_memory_impl(addr, size, exec);
2714   if (err != 0) {
2715     // the caller wants all commit errors to exit with the specified mesg:
2716     warn_fail_commit_memory(addr, size, exec, err);
2717     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2718   }
2719 }
2720 
2721 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2722 #ifndef MAP_HUGETLB
2723   #define MAP_HUGETLB 0x40000
2724 #endif
2725 
2726 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2727 #ifndef MADV_HUGEPAGE
2728   #define MADV_HUGEPAGE 14
2729 #endif
2730 
2731 int os::Linux::commit_memory_impl(char* addr, size_t size,
2732                                   size_t alignment_hint, bool exec) {
2733   int err = os::Linux::commit_memory_impl(addr, size, exec);
2734   if (err == 0) {
2735     realign_memory(addr, size, alignment_hint);
2736   }
2737   return err;
2738 }
2739 
2740 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2741                           bool exec) {
2742   return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2743 }
2744 
2745 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2746                                   size_t alignment_hint, bool exec,
2747                                   const char* mesg) {
2748   assert(mesg != NULL, "mesg must be specified");
2749   int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2750   if (err != 0) {
2751     // the caller wants all commit errors to exit with the specified mesg:
2752     warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2753     vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "%s", mesg);
2754   }
2755 }
2756 
2757 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2758   if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2759     // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2760     // be supported or the memory may already be backed by huge pages.
2761     ::madvise(addr, bytes, MADV_HUGEPAGE);
2762   }
2763 }
2764 
2765 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2766   // This method works by doing an mmap over an existing mmaping and effectively discarding
2767   // the existing pages. However it won't work for SHM-based large pages that cannot be
2768   // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2769   // small pages on top of the SHM segment. This method always works for small pages, so we
2770   // allow that in any case.
2771   if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2772     commit_memory(addr, bytes, alignment_hint, !ExecMem);
2773   }
2774 }
2775 
2776 void os::numa_make_global(char *addr, size_t bytes) {
2777   Linux::numa_interleave_memory(addr, bytes);
2778 }
2779 
2780 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2781 // bind policy to MPOL_PREFERRED for the current thread.
2782 #define USE_MPOL_PREFERRED 0
2783 
2784 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2785   // To make NUMA and large pages more robust when both enabled, we need to ease
2786   // the requirements on where the memory should be allocated. MPOL_BIND is the
2787   // default policy and it will force memory to be allocated on the specified
2788   // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2789   // the specified node, but will not force it. Using this policy will prevent
2790   // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2791   // free large pages.
2792   Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2793   Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2794 }
2795 
2796 bool os::numa_topology_changed() { return false; }
2797 
2798 size_t os::numa_get_groups_num() {
2799   // Return just the number of nodes in which it's possible to allocate memory
2800   // (in numa terminology, configured nodes).
2801   return Linux::numa_num_configured_nodes();
2802 }
2803 
2804 int os::numa_get_group_id() {
2805   int cpu_id = Linux::sched_getcpu();
2806   if (cpu_id != -1) {
2807     int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2808     if (lgrp_id != -1) {
2809       return lgrp_id;
2810     }
2811   }
2812   return 0;
2813 }
2814 
2815 int os::Linux::get_existing_num_nodes() {
2816   size_t node;
2817   size_t highest_node_number = Linux::numa_max_node();
2818   int num_nodes = 0;
2819 
2820   // Get the total number of nodes in the system including nodes without memory.
2821   for (node = 0; node <= highest_node_number; node++) {
2822     if (isnode_in_existing_nodes(node)) {
2823       num_nodes++;
2824     }
2825   }
2826   return num_nodes;
2827 }
2828 
2829 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2830   size_t highest_node_number = Linux::numa_max_node();
2831   size_t i = 0;
2832 
2833   // Map all node ids in which is possible to allocate memory. Also nodes are
2834   // not always consecutively available, i.e. available from 0 to the highest
2835   // node number.
2836   for (size_t node = 0; node <= highest_node_number; node++) {
2837     if (Linux::isnode_in_configured_nodes(node)) {
2838       ids[i++] = node;
2839     }
2840   }
2841   return i;
2842 }
2843 
2844 bool os::get_page_info(char *start, page_info* info) {
2845   return false;
2846 }
2847 
2848 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2849                      page_info* page_found) {
2850   return end;
2851 }
2852 
2853 
2854 int os::Linux::sched_getcpu_syscall(void) {
2855   unsigned int cpu = 0;
2856   int retval = -1;
2857 
2858 #if defined(IA32)
2859   #ifndef SYS_getcpu
2860     #define SYS_getcpu 318
2861   #endif
2862   retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2863 #elif defined(AMD64)
2864 // Unfortunately we have to bring all these macros here from vsyscall.h
2865 // to be able to compile on old linuxes.
2866   #define __NR_vgetcpu 2
2867   #define VSYSCALL_START (-10UL << 20)
2868   #define VSYSCALL_SIZE 1024
2869   #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2870   typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2871   vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2872   retval = vgetcpu(&cpu, NULL, NULL);
2873 #endif
2874 
2875   return (retval == -1) ? retval : cpu;
2876 }
2877 
2878 void os::Linux::sched_getcpu_init() {
2879   // sched_getcpu() should be in libc.
2880   set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2881                                   dlsym(RTLD_DEFAULT, "sched_getcpu")));
2882 
2883   // If it's not, try a direct syscall.
2884   if (sched_getcpu() == -1) {
2885     set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2886                                     (void*)&sched_getcpu_syscall));
2887   }
2888 }
2889 
2890 // Something to do with the numa-aware allocator needs these symbols
2891 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2892 extern "C" JNIEXPORT void numa_error(char *where) { }
2893 
2894 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2895 // load symbol from base version instead.
2896 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2897   void *f = dlvsym(handle, name, "libnuma_1.1");
2898   if (f == NULL) {
2899     f = dlsym(handle, name);
2900   }
2901   return f;
2902 }
2903 
2904 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2905 // Return NULL if the symbol is not defined in this particular version.
2906 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2907   return dlvsym(handle, name, "libnuma_1.2");
2908 }
2909 
2910 bool os::Linux::libnuma_init() {
2911   if (sched_getcpu() != -1) { // Requires sched_getcpu() support
2912     void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2913     if (handle != NULL) {
2914       set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2915                                            libnuma_dlsym(handle, "numa_node_to_cpus")));
2916       set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2917                                        libnuma_dlsym(handle, "numa_max_node")));
2918       set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2919                                                    libnuma_dlsym(handle, "numa_num_configured_nodes")));
2920       set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2921                                         libnuma_dlsym(handle, "numa_available")));
2922       set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2923                                             libnuma_dlsym(handle, "numa_tonode_memory")));
2924       set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2925                                                 libnuma_dlsym(handle, "numa_interleave_memory")));
2926       set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2927                                                 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2928       set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2929                                               libnuma_dlsym(handle, "numa_set_bind_policy")));
2930       set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2931                                                libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2932       set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2933                                        libnuma_dlsym(handle, "numa_distance")));
2934 
2935       if (numa_available() != -1) {
2936         set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2937         set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2938         set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2939         // Create an index -> node mapping, since nodes are not always consecutive
2940         _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2941         rebuild_nindex_to_node_map();
2942         // Create a cpu -> node mapping
2943         _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2944         rebuild_cpu_to_node_map();
2945         return true;
2946       }
2947     }
2948   }
2949   return false;
2950 }
2951 
2952 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {
2953   // Creating guard page is very expensive. Java thread has HotSpot
2954   // guard pages, only enable glibc guard page for non-Java threads.
2955   // (Remember: compiler thread is a Java thread, too!)
2956   return ((thr_type == java_thread || thr_type == compiler_thread) ? 0 : page_size());
2957 }
2958 
2959 void os::Linux::rebuild_nindex_to_node_map() {
2960   int highest_node_number = Linux::numa_max_node();
2961 
2962   nindex_to_node()->clear();
2963   for (int node = 0; node <= highest_node_number; node++) {
2964     if (Linux::isnode_in_existing_nodes(node)) {
2965       nindex_to_node()->append(node);
2966     }
2967   }
2968 }
2969 
2970 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2971 // The table is later used in get_node_by_cpu().
2972 void os::Linux::rebuild_cpu_to_node_map() {
2973   const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2974                               // in libnuma (possible values are starting from 16,
2975                               // and continuing up with every other power of 2, but less
2976                               // than the maximum number of CPUs supported by kernel), and
2977                               // is a subject to change (in libnuma version 2 the requirements
2978                               // are more reasonable) we'll just hardcode the number they use
2979                               // in the library.
2980   const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2981 
2982   size_t cpu_num = processor_count();
2983   size_t cpu_map_size = NCPUS / BitsPerCLong;
2984   size_t cpu_map_valid_size =
2985     MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2986 
2987   cpu_to_node()->clear();
2988   cpu_to_node()->at_grow(cpu_num - 1);
2989 
2990   size_t node_num = get_existing_num_nodes();
2991 
2992   int distance = 0;
2993   int closest_distance = INT_MAX;
2994   int closest_node = 0;
2995   unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2996   for (size_t i = 0; i < node_num; i++) {
2997     // Check if node is configured (not a memory-less node). If it is not, find
2998     // the closest configured node.
2999     if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3000       closest_distance = INT_MAX;
3001       // Check distance from all remaining nodes in the system. Ignore distance
3002       // from itself and from another non-configured node.
3003       for (size_t m = 0; m < node_num; m++) {
3004         if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3005           distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3006           // If a closest node is found, update. There is always at least one
3007           // configured node in the system so there is always at least one node
3008           // close.
3009           if (distance != 0 && distance < closest_distance) {
3010             closest_distance = distance;
3011             closest_node = nindex_to_node()->at(m);
3012           }
3013         }
3014       }
3015      } else {
3016        // Current node is already a configured node.
3017        closest_node = nindex_to_node()->at(i);
3018      }
3019 
3020     // Get cpus from the original node and map them to the closest node. If node
3021     // is a configured node (not a memory-less node), then original node and
3022     // closest node are the same.
3023     if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3024       for (size_t j = 0; j < cpu_map_valid_size; j++) {
3025         if (cpu_map[j] != 0) {
3026           for (size_t k = 0; k < BitsPerCLong; k++) {
3027             if (cpu_map[j] & (1UL << k)) {
3028               cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3029             }
3030           }
3031         }
3032       }
3033     }
3034   }
3035   FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
3036 }
3037 
3038 int os::Linux::get_node_by_cpu(int cpu_id) {
3039   if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3040     return cpu_to_node()->at(cpu_id);
3041   }
3042   return -1;
3043 }
3044 
3045 GrowableArray<int>* os::Linux::_cpu_to_node;
3046 GrowableArray<int>* os::Linux::_nindex_to_node;
3047 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3048 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3049 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3050 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3051 os::Linux::numa_available_func_t os::Linux::_numa_available;
3052 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3053 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3054 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3055 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3056 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3057 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3058 unsigned long* os::Linux::_numa_all_nodes;
3059 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3060 struct bitmask* os::Linux::_numa_nodes_ptr;
3061 
3062 bool os::pd_uncommit_memory(char* addr, size_t size) {
3063   uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3064                                      MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3065   return res  != (uintptr_t) MAP_FAILED;
3066 }
3067 
3068 static address get_stack_commited_bottom(address bottom, size_t size) {
3069   address nbot = bottom;
3070   address ntop = bottom + size;
3071 
3072   size_t page_sz = os::vm_page_size();
3073   unsigned pages = size / page_sz;
3074 
3075   unsigned char vec[1];
3076   unsigned imin = 1, imax = pages + 1, imid;
3077   int mincore_return_value = 0;
3078 
3079   assert(imin <= imax, "Unexpected page size");
3080 
3081   while (imin < imax) {
3082     imid = (imax + imin) / 2;
3083     nbot = ntop - (imid * page_sz);
3084 
3085     // Use a trick with mincore to check whether the page is mapped or not.
3086     // mincore sets vec to 1 if page resides in memory and to 0 if page
3087     // is swapped output but if page we are asking for is unmapped
3088     // it returns -1,ENOMEM
3089     mincore_return_value = mincore(nbot, page_sz, vec);
3090 
3091     if (mincore_return_value == -1) {
3092       // Page is not mapped go up
3093       // to find first mapped page
3094       if (errno != EAGAIN) {
3095         assert(errno == ENOMEM, "Unexpected mincore errno");
3096         imax = imid;
3097       }
3098     } else {
3099       // Page is mapped go down
3100       // to find first not mapped page
3101       imin = imid + 1;
3102     }
3103   }
3104 
3105   nbot = nbot + page_sz;
3106 
3107   // Adjust stack bottom one page up if last checked page is not mapped
3108   if (mincore_return_value == -1) {
3109     nbot = nbot + page_sz;
3110   }
3111 
3112   return nbot;
3113 }
3114 
3115 
3116 // Linux uses a growable mapping for the stack, and if the mapping for
3117 // the stack guard pages is not removed when we detach a thread the
3118 // stack cannot grow beyond the pages where the stack guard was
3119 // mapped.  If at some point later in the process the stack expands to
3120 // that point, the Linux kernel cannot expand the stack any further
3121 // because the guard pages are in the way, and a segfault occurs.
3122 //
3123 // However, it's essential not to split the stack region by unmapping
3124 // a region (leaving a hole) that's already part of the stack mapping,
3125 // so if the stack mapping has already grown beyond the guard pages at
3126 // the time we create them, we have to truncate the stack mapping.
3127 // So, we need to know the extent of the stack mapping when
3128 // create_stack_guard_pages() is called.
3129 
3130 // We only need this for stacks that are growable: at the time of
3131 // writing thread stacks don't use growable mappings (i.e. those
3132 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3133 // only applies to the main thread.
3134 
3135 // If the (growable) stack mapping already extends beyond the point
3136 // where we're going to put our guard pages, truncate the mapping at
3137 // that point by munmap()ping it.  This ensures that when we later
3138 // munmap() the guard pages we don't leave a hole in the stack
3139 // mapping. This only affects the main/initial thread
3140 
3141 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3142   if (os::Linux::is_initial_thread()) {
3143     // As we manually grow stack up to bottom inside create_attached_thread(),
3144     // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3145     // we don't need to do anything special.
3146     // Check it first, before calling heavy function.
3147     uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3148     unsigned char vec[1];
3149 
3150     if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3151       // Fallback to slow path on all errors, including EAGAIN
3152       stack_extent = (uintptr_t) get_stack_commited_bottom(
3153                                                            os::Linux::initial_thread_stack_bottom(),
3154                                                            (size_t)addr - stack_extent);
3155     }
3156 
3157     if (stack_extent < (uintptr_t)addr) {
3158       ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3159     }
3160   }
3161 
3162   return os::commit_memory(addr, size, !ExecMem);
3163 }
3164 
3165 // If this is a growable mapping, remove the guard pages entirely by
3166 // munmap()ping them.  If not, just call uncommit_memory(). This only
3167 // affects the main/initial thread, but guard against future OS changes
3168 // It's safe to always unmap guard pages for initial thread because we
3169 // always place it right after end of the mapped region
3170 
3171 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3172   uintptr_t stack_extent, stack_base;
3173 
3174   if (os::Linux::is_initial_thread()) {
3175     return ::munmap(addr, size) == 0;
3176   }
3177 
3178   return os::uncommit_memory(addr, size);
3179 }
3180 
3181 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3182 // at 'requested_addr'. If there are existing memory mappings at the same
3183 // location, however, they will be overwritten. If 'fixed' is false,
3184 // 'requested_addr' is only treated as a hint, the return value may or
3185 // may not start from the requested address. Unlike Linux mmap(), this
3186 // function returns NULL to indicate failure.
3187 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3188   char * addr;
3189   int flags;
3190 
3191   flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3192   if (fixed) {
3193     assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3194     flags |= MAP_FIXED;
3195   }
3196 
3197   // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3198   // touch an uncommitted page. Otherwise, the read/write might
3199   // succeed if we have enough swap space to back the physical page.
3200   addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3201                        flags, -1, 0);
3202 
3203   return addr == MAP_FAILED ? NULL : addr;
3204 }
3205 
3206 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3207 //   (req_addr != NULL) or with a given alignment.
3208 //  - bytes shall be a multiple of alignment.
3209 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3210 //  - alignment sets the alignment at which memory shall be allocated.
3211 //     It must be a multiple of allocation granularity.
3212 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3213 //  req_addr or NULL.
3214 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3215 
3216   size_t extra_size = bytes;
3217   if (req_addr == NULL && alignment > 0) {
3218     extra_size += alignment;
3219   }
3220 
3221   char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3222     MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3223     -1, 0);
3224   if (start == MAP_FAILED) {
3225     start = NULL;
3226   } else {
3227     if (req_addr != NULL) {
3228       if (start != req_addr) {
3229         ::munmap(start, extra_size);
3230         start = NULL;
3231       }
3232     } else {
3233       char* const start_aligned = align_up(start, alignment);
3234       char* const end_aligned = start_aligned + bytes;
3235       char* const end = start + extra_size;
3236       if (start_aligned > start) {
3237         ::munmap(start, start_aligned - start);
3238       }
3239       if (end_aligned < end) {
3240         ::munmap(end_aligned, end - end_aligned);
3241       }
3242       start = start_aligned;
3243     }
3244   }
3245   return start;
3246 }
3247 
3248 static int anon_munmap(char * addr, size_t size) {
3249   return ::munmap(addr, size) == 0;
3250 }
3251 
3252 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3253                             size_t alignment_hint) {
3254   return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3255 }
3256 
3257 bool os::pd_release_memory(char* addr, size_t size) {
3258   return anon_munmap(addr, size);
3259 }
3260 
3261 static bool linux_mprotect(char* addr, size_t size, int prot) {
3262   // Linux wants the mprotect address argument to be page aligned.
3263   char* bottom = (char*)align_down((intptr_t)addr, os::Linux::page_size());
3264 
3265   // According to SUSv3, mprotect() should only be used with mappings
3266   // established by mmap(), and mmap() always maps whole pages. Unaligned
3267   // 'addr' likely indicates problem in the VM (e.g. trying to change
3268   // protection of malloc'ed or statically allocated memory). Check the
3269   // caller if you hit this assert.
3270   assert(addr == bottom, "sanity check");
3271 
3272   size = align_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3273   return ::mprotect(bottom, size, prot) == 0;
3274 }
3275 
3276 // Set protections specified
3277 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3278                         bool is_committed) {
3279   unsigned int p = 0;
3280   switch (prot) {
3281   case MEM_PROT_NONE: p = PROT_NONE; break;
3282   case MEM_PROT_READ: p = PROT_READ; break;
3283   case MEM_PROT_RW:   p = PROT_READ|PROT_WRITE; break;
3284   case MEM_PROT_RWX:  p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3285   default:
3286     ShouldNotReachHere();
3287   }
3288   // is_committed is unused.
3289   return linux_mprotect(addr, bytes, p);
3290 }
3291 
3292 bool os::guard_memory(char* addr, size_t size) {
3293   return linux_mprotect(addr, size, PROT_NONE);
3294 }
3295 
3296 bool os::unguard_memory(char* addr, size_t size) {
3297   return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3298 }
3299 
3300 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3301                                                     size_t page_size) {
3302   bool result = false;
3303   void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3304                  MAP_ANONYMOUS|MAP_PRIVATE,
3305                  -1, 0);
3306   if (p != MAP_FAILED) {
3307     void *aligned_p = align_up(p, page_size);
3308 
3309     result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3310 
3311     munmap(p, page_size * 2);
3312   }
3313 
3314   if (warn && !result) {
3315     warning("TransparentHugePages is not supported by the operating system.");
3316   }
3317 
3318   return result;
3319 }
3320 
3321 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3322   bool result = false;
3323   void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3324                  MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3325                  -1, 0);
3326 
3327   if (p != MAP_FAILED) {
3328     // We don't know if this really is a huge page or not.
3329     FILE *fp = fopen("/proc/self/maps", "r");
3330     if (fp) {
3331       while (!feof(fp)) {
3332         char chars[257];
3333         long x = 0;
3334         if (fgets(chars, sizeof(chars), fp)) {
3335           if (sscanf(chars, "%lx-%*x", &x) == 1
3336               && x == (long)p) {
3337             if (strstr (chars, "hugepage")) {
3338               result = true;
3339               break;
3340             }
3341           }
3342         }
3343       }
3344       fclose(fp);
3345     }
3346     munmap(p, page_size);
3347   }
3348 
3349   if (warn && !result) {
3350     warning("HugeTLBFS is not supported by the operating system.");
3351   }
3352 
3353   return result;
3354 }
3355 
3356 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3357 //
3358 // From the coredump_filter documentation:
3359 //
3360 // - (bit 0) anonymous private memory
3361 // - (bit 1) anonymous shared memory
3362 // - (bit 2) file-backed private memory
3363 // - (bit 3) file-backed shared memory
3364 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3365 //           effective only if the bit 2 is cleared)
3366 // - (bit 5) hugetlb private memory
3367 // - (bit 6) hugetlb shared memory
3368 //
3369 static void set_coredump_filter(void) {
3370   FILE *f;
3371   long cdm;
3372 
3373   if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3374     return;
3375   }
3376 
3377   if (fscanf(f, "%lx", &cdm) != 1) {
3378     fclose(f);
3379     return;
3380   }
3381 
3382   rewind(f);
3383 
3384   if ((cdm & LARGEPAGES_BIT) == 0) {
3385     cdm |= LARGEPAGES_BIT;
3386     fprintf(f, "%#lx", cdm);
3387   }
3388 
3389   fclose(f);
3390 }
3391 
3392 // Large page support
3393 
3394 static size_t _large_page_size = 0;
3395 
3396 size_t os::Linux::find_large_page_size() {
3397   size_t large_page_size = 0;
3398 
3399   // large_page_size on Linux is used to round up heap size. x86 uses either
3400   // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3401   // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3402   // page as large as 256M.
3403   //
3404   // Here we try to figure out page size by parsing /proc/meminfo and looking
3405   // for a line with the following format:
3406   //    Hugepagesize:     2048 kB
3407   //
3408   // If we can't determine the value (e.g. /proc is not mounted, or the text
3409   // format has been changed), we'll use the largest page size supported by
3410   // the processor.
3411 
3412 #ifndef ZERO
3413   large_page_size =
3414     AARCH64_ONLY(2 * M)
3415     AMD64_ONLY(2 * M)
3416     ARM32_ONLY(2 * M)
3417     IA32_ONLY(4 * M)
3418     IA64_ONLY(256 * M)
3419     PPC_ONLY(4 * M)
3420     S390_ONLY(1 * M)
3421     SPARC_ONLY(4 * M);
3422 #endif // ZERO
3423 
3424   FILE *fp = fopen("/proc/meminfo", "r");
3425   if (fp) {
3426     while (!feof(fp)) {
3427       int x = 0;
3428       char buf[16];
3429       if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3430         if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3431           large_page_size = x * K;
3432           break;
3433         }
3434       } else {
3435         // skip to next line
3436         for (;;) {
3437           int ch = fgetc(fp);
3438           if (ch == EOF || ch == (int)'\n') break;
3439         }
3440       }
3441     }
3442     fclose(fp);
3443   }
3444 
3445   if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3446     warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3447             SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3448             proper_unit_for_byte_size(large_page_size));
3449   }
3450 
3451   return large_page_size;
3452 }
3453 
3454 size_t os::Linux::setup_large_page_size() {
3455   _large_page_size = Linux::find_large_page_size();
3456   const size_t default_page_size = (size_t)Linux::page_size();
3457   if (_large_page_size > default_page_size) {
3458     _page_sizes[0] = _large_page_size;
3459     _page_sizes[1] = default_page_size;
3460     _page_sizes[2] = 0;
3461   }
3462 
3463   return _large_page_size;
3464 }
3465 
3466 bool os::Linux::setup_large_page_type(size_t page_size) {
3467   if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3468       FLAG_IS_DEFAULT(UseSHM) &&
3469       FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3470 
3471     // The type of large pages has not been specified by the user.
3472 
3473     // Try UseHugeTLBFS and then UseSHM.
3474     UseHugeTLBFS = UseSHM = true;
3475 
3476     // Don't try UseTransparentHugePages since there are known
3477     // performance issues with it turned on. This might change in the future.
3478     UseTransparentHugePages = false;
3479   }
3480 
3481   if (UseTransparentHugePages) {
3482     bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3483     if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3484       UseHugeTLBFS = false;
3485       UseSHM = false;
3486       return true;
3487     }
3488     UseTransparentHugePages = false;
3489   }
3490 
3491   if (UseHugeTLBFS) {
3492     bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3493     if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3494       UseSHM = false;
3495       return true;
3496     }
3497     UseHugeTLBFS = false;
3498   }
3499 
3500   return UseSHM;
3501 }
3502 
3503 void os::large_page_init() {
3504   if (!UseLargePages &&
3505       !UseTransparentHugePages &&
3506       !UseHugeTLBFS &&
3507       !UseSHM) {
3508     // Not using large pages.
3509     return;
3510   }
3511 
3512   if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3513     // The user explicitly turned off large pages.
3514     // Ignore the rest of the large pages flags.
3515     UseTransparentHugePages = false;
3516     UseHugeTLBFS = false;
3517     UseSHM = false;
3518     return;
3519   }
3520 
3521   size_t large_page_size = Linux::setup_large_page_size();
3522   UseLargePages          = Linux::setup_large_page_type(large_page_size);
3523 
3524   set_coredump_filter();
3525 }
3526 
3527 #ifndef SHM_HUGETLB
3528   #define SHM_HUGETLB 04000
3529 #endif
3530 
3531 #define shm_warning_format(format, ...)              \
3532   do {                                               \
3533     if (UseLargePages &&                             \
3534         (!FLAG_IS_DEFAULT(UseLargePages) ||          \
3535          !FLAG_IS_DEFAULT(UseSHM) ||                 \
3536          !FLAG_IS_DEFAULT(LargePageSizeInBytes))) {  \
3537       warning(format, __VA_ARGS__);                  \
3538     }                                                \
3539   } while (0)
3540 
3541 #define shm_warning(str) shm_warning_format("%s", str)
3542 
3543 #define shm_warning_with_errno(str)                \
3544   do {                                             \
3545     int err = errno;                               \
3546     shm_warning_format(str " (error = %d)", err);  \
3547   } while (0)
3548 
3549 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3550   assert(is_aligned(bytes, alignment), "Must be divisible by the alignment");
3551 
3552   if (!is_aligned(alignment, SHMLBA)) {
3553     assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3554     return NULL;
3555   }
3556 
3557   // To ensure that we get 'alignment' aligned memory from shmat,
3558   // we pre-reserve aligned virtual memory and then attach to that.
3559 
3560   char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3561   if (pre_reserved_addr == NULL) {
3562     // Couldn't pre-reserve aligned memory.
3563     shm_warning("Failed to pre-reserve aligned memory for shmat.");
3564     return NULL;
3565   }
3566 
3567   // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3568   char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3569 
3570   if ((intptr_t)addr == -1) {
3571     int err = errno;
3572     shm_warning_with_errno("Failed to attach shared memory.");
3573 
3574     assert(err != EACCES, "Unexpected error");
3575     assert(err != EIDRM,  "Unexpected error");
3576     assert(err != EINVAL, "Unexpected error");
3577 
3578     // Since we don't know if the kernel unmapped the pre-reserved memory area
3579     // we can't unmap it, since that would potentially unmap memory that was
3580     // mapped from other threads.
3581     return NULL;
3582   }
3583 
3584   return addr;
3585 }
3586 
3587 static char* shmat_at_address(int shmid, char* req_addr) {
3588   if (!is_aligned(req_addr, SHMLBA)) {
3589     assert(false, "Requested address needs to be SHMLBA aligned");
3590     return NULL;
3591   }
3592 
3593   char* addr = (char*)shmat(shmid, req_addr, 0);
3594 
3595   if ((intptr_t)addr == -1) {
3596     shm_warning_with_errno("Failed to attach shared memory.");
3597     return NULL;
3598   }
3599 
3600   return addr;
3601 }
3602 
3603 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3604   // If a req_addr has been provided, we assume that the caller has already aligned the address.
3605   if (req_addr != NULL) {
3606     assert(is_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3607     assert(is_aligned(req_addr, alignment), "Must be divisible by given alignment");
3608     return shmat_at_address(shmid, req_addr);
3609   }
3610 
3611   // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3612   // return large page size aligned memory addresses when req_addr == NULL.
3613   // However, if the alignment is larger than the large page size, we have
3614   // to manually ensure that the memory returned is 'alignment' aligned.
3615   if (alignment > os::large_page_size()) {
3616     assert(is_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3617     return shmat_with_alignment(shmid, bytes, alignment);
3618   } else {
3619     return shmat_at_address(shmid, NULL);
3620   }
3621 }
3622 
3623 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3624                                             char* req_addr, bool exec) {
3625   // "exec" is passed in but not used.  Creating the shared image for
3626   // the code cache doesn't have an SHM_X executable permission to check.
3627   assert(UseLargePages && UseSHM, "only for SHM large pages");
3628   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3629   assert(is_aligned(req_addr, alignment), "Unaligned address");
3630 
3631   if (!is_aligned(bytes, os::large_page_size())) {
3632     return NULL; // Fallback to small pages.
3633   }
3634 
3635   // Create a large shared memory region to attach to based on size.
3636   // Currently, size is the total size of the heap.
3637   int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3638   if (shmid == -1) {
3639     // Possible reasons for shmget failure:
3640     // 1. shmmax is too small for Java heap.
3641     //    > check shmmax value: cat /proc/sys/kernel/shmmax
3642     //    > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3643     // 2. not enough large page memory.
3644     //    > check available large pages: cat /proc/meminfo
3645     //    > increase amount of large pages:
3646     //          echo new_value > /proc/sys/vm/nr_hugepages
3647     //      Note 1: different Linux may use different name for this property,
3648     //            e.g. on Redhat AS-3 it is "hugetlb_pool".
3649     //      Note 2: it's possible there's enough physical memory available but
3650     //            they are so fragmented after a long run that they can't
3651     //            coalesce into large pages. Try to reserve large pages when
3652     //            the system is still "fresh".
3653     shm_warning_with_errno("Failed to reserve shared memory.");
3654     return NULL;
3655   }
3656 
3657   // Attach to the region.
3658   char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3659 
3660   // Remove shmid. If shmat() is successful, the actual shared memory segment
3661   // will be deleted when it's detached by shmdt() or when the process
3662   // terminates. If shmat() is not successful this will remove the shared
3663   // segment immediately.
3664   shmctl(shmid, IPC_RMID, NULL);
3665 
3666   return addr;
3667 }
3668 
3669 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3670                                         int error) {
3671   assert(error == ENOMEM, "Only expect to fail if no memory is available");
3672 
3673   bool warn_on_failure = UseLargePages &&
3674       (!FLAG_IS_DEFAULT(UseLargePages) ||
3675        !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3676        !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3677 
3678   if (warn_on_failure) {
3679     char msg[128];
3680     jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3681                  PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3682     warning("%s", msg);
3683   }
3684 }
3685 
3686 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3687                                                         char* req_addr,
3688                                                         bool exec) {
3689   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3690   assert(is_aligned(bytes, os::large_page_size()), "Unaligned size");
3691   assert(is_aligned(req_addr, os::large_page_size()), "Unaligned address");
3692 
3693   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3694   char* addr = (char*)::mmap(req_addr, bytes, prot,
3695                              MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3696                              -1, 0);
3697 
3698   if (addr == MAP_FAILED) {
3699     warn_on_large_pages_failure(req_addr, bytes, errno);
3700     return NULL;
3701   }
3702 
3703   assert(is_aligned(addr, os::large_page_size()), "Must be");
3704 
3705   return addr;
3706 }
3707 
3708 // Reserve memory using mmap(MAP_HUGETLB).
3709 //  - bytes shall be a multiple of alignment.
3710 //  - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3711 //  - alignment sets the alignment at which memory shall be allocated.
3712 //     It must be a multiple of allocation granularity.
3713 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3714 //  req_addr or NULL.
3715 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3716                                                          size_t alignment,
3717                                                          char* req_addr,
3718                                                          bool exec) {
3719   size_t large_page_size = os::large_page_size();
3720   assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3721 
3722   assert(is_aligned(req_addr, alignment), "Must be");
3723   assert(is_aligned(bytes, alignment), "Must be");
3724 
3725   // First reserve - but not commit - the address range in small pages.
3726   char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3727 
3728   if (start == NULL) {
3729     return NULL;
3730   }
3731 
3732   assert(is_aligned(start, alignment), "Must be");
3733 
3734   char* end = start + bytes;
3735 
3736   // Find the regions of the allocated chunk that can be promoted to large pages.
3737   char* lp_start = align_up(start, large_page_size);
3738   char* lp_end   = align_down(end, large_page_size);
3739 
3740   size_t lp_bytes = lp_end - lp_start;
3741 
3742   assert(is_aligned(lp_bytes, large_page_size), "Must be");
3743 
3744   if (lp_bytes == 0) {
3745     // The mapped region doesn't even span the start and the end of a large page.
3746     // Fall back to allocate a non-special area.
3747     ::munmap(start, end - start);
3748     return NULL;
3749   }
3750 
3751   int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3752 
3753   void* result;
3754 
3755   // Commit small-paged leading area.
3756   if (start != lp_start) {
3757     result = ::mmap(start, lp_start - start, prot,
3758                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3759                     -1, 0);
3760     if (result == MAP_FAILED) {
3761       ::munmap(lp_start, end - lp_start);
3762       return NULL;
3763     }
3764   }
3765 
3766   // Commit large-paged area.
3767   result = ::mmap(lp_start, lp_bytes, prot,
3768                   MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3769                   -1, 0);
3770   if (result == MAP_FAILED) {
3771     warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3772     // If the mmap above fails, the large pages region will be unmapped and we
3773     // have regions before and after with small pages. Release these regions.
3774     //
3775     // |  mapped  |  unmapped  |  mapped  |
3776     // ^          ^            ^          ^
3777     // start      lp_start     lp_end     end
3778     //
3779     ::munmap(start, lp_start - start);
3780     ::munmap(lp_end, end - lp_end);
3781     return NULL;
3782   }
3783 
3784   // Commit small-paged trailing area.
3785   if (lp_end != end) {
3786     result = ::mmap(lp_end, end - lp_end, prot,
3787                     MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3788                     -1, 0);
3789     if (result == MAP_FAILED) {
3790       ::munmap(start, lp_end - start);
3791       return NULL;
3792     }
3793   }
3794 
3795   return start;
3796 }
3797 
3798 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3799                                                    size_t alignment,
3800                                                    char* req_addr,
3801                                                    bool exec) {
3802   assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3803   assert(is_aligned(req_addr, alignment), "Must be");
3804   assert(is_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3805   assert(is_power_of_2(os::large_page_size()), "Must be");
3806   assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3807 
3808   if (is_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3809     return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3810   } else {
3811     return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3812   }
3813 }
3814 
3815 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3816                                  char* req_addr, bool exec) {
3817   assert(UseLargePages, "only for large pages");
3818 
3819   char* addr;
3820   if (UseSHM) {
3821     addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3822   } else {
3823     assert(UseHugeTLBFS, "must be");
3824     addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3825   }
3826 
3827   if (addr != NULL) {
3828     if (UseNUMAInterleaving) {
3829       numa_make_global(addr, bytes);
3830     }
3831 
3832     // The memory is committed
3833     MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3834   }
3835 
3836   return addr;
3837 }
3838 
3839 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3840   // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3841   return shmdt(base) == 0;
3842 }
3843 
3844 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3845   return pd_release_memory(base, bytes);
3846 }
3847 
3848 bool os::release_memory_special(char* base, size_t bytes) {
3849   bool res;
3850   if (MemTracker::tracking_level() > NMT_minimal) {
3851     Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3852     res = os::Linux::release_memory_special_impl(base, bytes);
3853     if (res) {
3854       tkr.record((address)base, bytes);
3855     }
3856 
3857   } else {
3858     res = os::Linux::release_memory_special_impl(base, bytes);
3859   }
3860   return res;
3861 }
3862 
3863 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3864   assert(UseLargePages, "only for large pages");
3865   bool res;
3866 
3867   if (UseSHM) {
3868     res = os::Linux::release_memory_special_shm(base, bytes);
3869   } else {
3870     assert(UseHugeTLBFS, "must be");
3871     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3872   }
3873   return res;
3874 }
3875 
3876 size_t os::large_page_size() {
3877   return _large_page_size;
3878 }
3879 
3880 // With SysV SHM the entire memory region must be allocated as shared
3881 // memory.
3882 // HugeTLBFS allows application to commit large page memory on demand.
3883 // However, when committing memory with HugeTLBFS fails, the region
3884 // that was supposed to be committed will lose the old reservation
3885 // and allow other threads to steal that memory region. Because of this
3886 // behavior we can't commit HugeTLBFS memory.
3887 bool os::can_commit_large_page_memory() {
3888   return UseTransparentHugePages;
3889 }
3890 
3891 bool os::can_execute_large_page_memory() {
3892   return UseTransparentHugePages || UseHugeTLBFS;
3893 }
3894 
3895 // Reserve memory at an arbitrary address, only if that area is
3896 // available (and not reserved for something else).
3897 
3898 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3899   const int max_tries = 10;
3900   char* base[max_tries];
3901   size_t size[max_tries];
3902   const size_t gap = 0x000000;
3903 
3904   // Assert only that the size is a multiple of the page size, since
3905   // that's all that mmap requires, and since that's all we really know
3906   // about at this low abstraction level.  If we need higher alignment,
3907   // we can either pass an alignment to this method or verify alignment
3908   // in one of the methods further up the call chain.  See bug 5044738.
3909   assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3910 
3911   // Repeatedly allocate blocks until the block is allocated at the
3912   // right spot.
3913 
3914   // Linux mmap allows caller to pass an address as hint; give it a try first,
3915   // if kernel honors the hint then we can return immediately.
3916   char * addr = anon_mmap(requested_addr, bytes, false);
3917   if (addr == requested_addr) {
3918     return requested_addr;
3919   }
3920 
3921   if (addr != NULL) {
3922     // mmap() is successful but it fails to reserve at the requested address
3923     anon_munmap(addr, bytes);
3924   }
3925 
3926   int i;
3927   for (i = 0; i < max_tries; ++i) {
3928     base[i] = reserve_memory(bytes);
3929 
3930     if (base[i] != NULL) {
3931       // Is this the block we wanted?
3932       if (base[i] == requested_addr) {
3933         size[i] = bytes;
3934         break;
3935       }
3936 
3937       // Does this overlap the block we wanted? Give back the overlapped
3938       // parts and try again.
3939 
3940       ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3941       if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3942         unmap_memory(base[i], top_overlap);
3943         base[i] += top_overlap;
3944         size[i] = bytes - top_overlap;
3945       } else {
3946         ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3947         if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3948           unmap_memory(requested_addr, bottom_overlap);
3949           size[i] = bytes - bottom_overlap;
3950         } else {
3951           size[i] = bytes;
3952         }
3953       }
3954     }
3955   }
3956 
3957   // Give back the unused reserved pieces.
3958 
3959   for (int j = 0; j < i; ++j) {
3960     if (base[j] != NULL) {
3961       unmap_memory(base[j], size[j]);
3962     }
3963   }
3964 
3965   if (i < max_tries) {
3966     return requested_addr;
3967   } else {
3968     return NULL;
3969   }
3970 }
3971 
3972 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3973   return ::read(fd, buf, nBytes);
3974 }
3975 
3976 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3977   return ::pread(fd, buf, nBytes, offset);
3978 }
3979 
3980 // Short sleep, direct OS call.
3981 //
3982 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3983 // sched_yield(2) will actually give up the CPU:
3984 //
3985 //   * Alone on this pariticular CPU, keeps running.
3986 //   * Before the introduction of "skip_buddy" with "compat_yield" disabled
3987 //     (pre 2.6.39).
3988 //
3989 // So calling this with 0 is an alternative.
3990 //
3991 void os::naked_short_sleep(jlong ms) {
3992   struct timespec req;
3993 
3994   assert(ms < 1000, "Un-interruptable sleep, short time use only");
3995   req.tv_sec = 0;
3996   if (ms > 0) {
3997     req.tv_nsec = (ms % 1000) * 1000000;
3998   } else {
3999     req.tv_nsec = 1;
4000   }
4001 
4002   nanosleep(&req, NULL);
4003 
4004   return;
4005 }
4006 
4007 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4008 void os::infinite_sleep() {
4009   while (true) {    // sleep forever ...
4010     ::sleep(100);   // ... 100 seconds at a time
4011   }
4012 }
4013 
4014 // Used to convert frequent JVM_Yield() to nops
4015 bool os::dont_yield() {
4016   return DontYieldALot;
4017 }
4018 
4019 void os::naked_yield() {
4020   sched_yield();
4021 }
4022 
4023 ////////////////////////////////////////////////////////////////////////////////
4024 // thread priority support
4025 
4026 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4027 // only supports dynamic priority, static priority must be zero. For real-time
4028 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4029 // However, for large multi-threaded applications, SCHED_RR is not only slower
4030 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4031 // of 5 runs - Sep 2005).
4032 //
4033 // The following code actually changes the niceness of kernel-thread/LWP. It
4034 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4035 // not the entire user process, and user level threads are 1:1 mapped to kernel
4036 // threads. It has always been the case, but could change in the future. For
4037 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4038 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4039 
4040 int os::java_to_os_priority[CriticalPriority + 1] = {
4041   19,              // 0 Entry should never be used
4042 
4043    4,              // 1 MinPriority
4044    3,              // 2
4045    2,              // 3
4046 
4047    1,              // 4
4048    0,              // 5 NormPriority
4049   -1,              // 6
4050 
4051   -2,              // 7
4052   -3,              // 8
4053   -4,              // 9 NearMaxPriority
4054 
4055   -5,              // 10 MaxPriority
4056 
4057   -5               // 11 CriticalPriority
4058 };
4059 
4060 static int prio_init() {
4061   if (ThreadPriorityPolicy == 1) {
4062     // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4063     // if effective uid is not root. Perhaps, a more elegant way of doing
4064     // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4065     if (geteuid() != 0) {
4066       if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4067         warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4068       }
4069       ThreadPriorityPolicy = 0;
4070     }
4071   }
4072   if (UseCriticalJavaThreadPriority) {
4073     os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4074   }
4075   return 0;
4076 }
4077 
4078 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4079   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
4080 
4081   int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4082   return (ret == 0) ? OS_OK : OS_ERR;
4083 }
4084 
4085 OSReturn os::get_native_priority(const Thread* const thread,
4086                                  int *priority_ptr) {
4087   if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
4088     *priority_ptr = java_to_os_priority[NormPriority];
4089     return OS_OK;
4090   }
4091 
4092   errno = 0;
4093   *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4094   return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4095 }
4096 
4097 // Hint to the underlying OS that a task switch would not be good.
4098 // Void return because it's a hint and can fail.
4099 void os::hint_no_preempt() {}
4100 
4101 ////////////////////////////////////////////////////////////////////////////////
4102 // suspend/resume support
4103 
4104 //  The low-level signal-based suspend/resume support is a remnant from the
4105 //  old VM-suspension that used to be for java-suspension, safepoints etc,
4106 //  within hotspot. Currently used by JFR's OSThreadSampler
4107 //
4108 //  The remaining code is greatly simplified from the more general suspension
4109 //  code that used to be used.
4110 //
4111 //  The protocol is quite simple:
4112 //  - suspend:
4113 //      - sends a signal to the target thread
4114 //      - polls the suspend state of the osthread using a yield loop
4115 //      - target thread signal handler (SR_handler) sets suspend state
4116 //        and blocks in sigsuspend until continued
4117 //  - resume:
4118 //      - sets target osthread state to continue
4119 //      - sends signal to end the sigsuspend loop in the SR_handler
4120 //
4121 //  Note that the SR_lock plays no role in this suspend/resume protocol,
4122 //  but is checked for NULL in SR_handler as a thread termination indicator.
4123 //  The SR_lock is, however, used by JavaThread::java_suspend()/java_resume() APIs.
4124 //
4125 //  Note that resume_clear_context() and suspend_save_context() are needed
4126 //  by SR_handler(), so that fetch_frame_from_ucontext() works,
4127 //  which in part is used by:
4128 //    - Forte Analyzer: AsyncGetCallTrace()
4129 //    - StackBanging: get_frame_at_stack_banging_point()
4130 
4131 static void resume_clear_context(OSThread *osthread) {
4132   osthread->set_ucontext(NULL);
4133   osthread->set_siginfo(NULL);
4134 }
4135 
4136 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
4137                                  ucontext_t* context) {
4138   osthread->set_ucontext(context);
4139   osthread->set_siginfo(siginfo);
4140 }
4141 
4142 // Handler function invoked when a thread's execution is suspended or
4143 // resumed. We have to be careful that only async-safe functions are
4144 // called here (Note: most pthread functions are not async safe and
4145 // should be avoided.)
4146 //
4147 // Note: sigwait() is a more natural fit than sigsuspend() from an
4148 // interface point of view, but sigwait() prevents the signal hander
4149 // from being run. libpthread would get very confused by not having
4150 // its signal handlers run and prevents sigwait()'s use with the
4151 // mutex granting granting signal.
4152 //
4153 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4154 //
4155 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4156   // Save and restore errno to avoid confusing native code with EINTR
4157   // after sigsuspend.
4158   int old_errno = errno;
4159 
4160   Thread* thread = Thread::current_or_null_safe();
4161   assert(thread != NULL, "Missing current thread in SR_handler");
4162 
4163   // On some systems we have seen signal delivery get "stuck" until the signal
4164   // mask is changed as part of thread termination. Check that the current thread
4165   // has not already terminated (via SR_lock()) - else the following assertion
4166   // will fail because the thread is no longer a JavaThread as the ~JavaThread
4167   // destructor has completed.
4168 
4169   if (thread->SR_lock() == NULL) {
4170     return;
4171   }
4172 
4173   assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4174 
4175   OSThread* osthread = thread->osthread();
4176 
4177   os::SuspendResume::State current = osthread->sr.state();
4178   if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4179     suspend_save_context(osthread, siginfo, context);
4180 
4181     // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4182     os::SuspendResume::State state = osthread->sr.suspended();
4183     if (state == os::SuspendResume::SR_SUSPENDED) {
4184       sigset_t suspend_set;  // signals for sigsuspend()
4185       sigemptyset(&suspend_set);
4186       // get current set of blocked signals and unblock resume signal
4187       pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4188       sigdelset(&suspend_set, SR_signum);
4189 
4190       sr_semaphore.signal();
4191       // wait here until we are resumed
4192       while (1) {
4193         sigsuspend(&suspend_set);
4194 
4195         os::SuspendResume::State result = osthread->sr.running();
4196         if (result == os::SuspendResume::SR_RUNNING) {
4197           sr_semaphore.signal();
4198           break;
4199         }
4200       }
4201 
4202     } else if (state == os::SuspendResume::SR_RUNNING) {
4203       // request was cancelled, continue
4204     } else {
4205       ShouldNotReachHere();
4206     }
4207 
4208     resume_clear_context(osthread);
4209   } else if (current == os::SuspendResume::SR_RUNNING) {
4210     // request was cancelled, continue
4211   } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4212     // ignore
4213   } else {
4214     // ignore
4215   }
4216 
4217   errno = old_errno;
4218 }
4219 
4220 static int SR_initialize() {
4221   struct sigaction act;
4222   char *s;
4223 
4224   // Get signal number to use for suspend/resume
4225   if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4226     int sig = ::strtol(s, 0, 10);
4227     if (sig > MAX2(SIGSEGV, SIGBUS) &&  // See 4355769.
4228         sig < NSIG) {                   // Must be legal signal and fit into sigflags[].
4229       SR_signum = sig;
4230     } else {
4231       warning("You set _JAVA_SR_SIGNUM=%d. It must be in range [%d, %d]. Using %d instead.",
4232               sig, MAX2(SIGSEGV, SIGBUS)+1, NSIG-1, SR_signum);
4233     }
4234   }
4235 
4236   assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4237          "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4238 
4239   sigemptyset(&SR_sigset);
4240   sigaddset(&SR_sigset, SR_signum);
4241 
4242   // Set up signal handler for suspend/resume
4243   act.sa_flags = SA_RESTART|SA_SIGINFO;
4244   act.sa_handler = (void (*)(int)) SR_handler;
4245 
4246   // SR_signum is blocked by default.
4247   // 4528190 - We also need to block pthread restart signal (32 on all
4248   // supported Linux platforms). Note that LinuxThreads need to block
4249   // this signal for all threads to work properly. So we don't have
4250   // to use hard-coded signal number when setting up the mask.
4251   pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4252 
4253   if (sigaction(SR_signum, &act, 0) == -1) {
4254     return -1;
4255   }
4256 
4257   // Save signal flag
4258   os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4259   return 0;
4260 }
4261 
4262 static int sr_notify(OSThread* osthread) {
4263   int status = pthread_kill(osthread->pthread_id(), SR_signum);
4264   assert_status(status == 0, status, "pthread_kill");
4265   return status;
4266 }
4267 
4268 // "Randomly" selected value for how long we want to spin
4269 // before bailing out on suspending a thread, also how often
4270 // we send a signal to a thread we want to resume
4271 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4272 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4273 
4274 // returns true on success and false on error - really an error is fatal
4275 // but this seems the normal response to library errors
4276 static bool do_suspend(OSThread* osthread) {
4277   assert(osthread->sr.is_running(), "thread should be running");
4278   assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4279 
4280   // mark as suspended and send signal
4281   if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4282     // failed to switch, state wasn't running?
4283     ShouldNotReachHere();
4284     return false;
4285   }
4286 
4287   if (sr_notify(osthread) != 0) {
4288     ShouldNotReachHere();
4289   }
4290 
4291   // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4292   while (true) {
4293     if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4294       break;
4295     } else {
4296       // timeout
4297       os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4298       if (cancelled == os::SuspendResume::SR_RUNNING) {
4299         return false;
4300       } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4301         // make sure that we consume the signal on the semaphore as well
4302         sr_semaphore.wait();
4303         break;
4304       } else {
4305         ShouldNotReachHere();
4306         return false;
4307       }
4308     }
4309   }
4310 
4311   guarantee(osthread->sr.is_suspended(), "Must be suspended");
4312   return true;
4313 }
4314 
4315 static void do_resume(OSThread* osthread) {
4316   assert(osthread->sr.is_suspended(), "thread should be suspended");
4317   assert(!sr_semaphore.trywait(), "invalid semaphore state");
4318 
4319   if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4320     // failed to switch to WAKEUP_REQUEST
4321     ShouldNotReachHere();
4322     return;
4323   }
4324 
4325   while (true) {
4326     if (sr_notify(osthread) == 0) {
4327       if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4328         if (osthread->sr.is_running()) {
4329           return;
4330         }
4331       }
4332     } else {
4333       ShouldNotReachHere();
4334     }
4335   }
4336 
4337   guarantee(osthread->sr.is_running(), "Must be running!");
4338 }
4339 
4340 ///////////////////////////////////////////////////////////////////////////////////
4341 // signal handling (except suspend/resume)
4342 
4343 // This routine may be used by user applications as a "hook" to catch signals.
4344 // The user-defined signal handler must pass unrecognized signals to this
4345 // routine, and if it returns true (non-zero), then the signal handler must
4346 // return immediately.  If the flag "abort_if_unrecognized" is true, then this
4347 // routine will never retun false (zero), but instead will execute a VM panic
4348 // routine kill the process.
4349 //
4350 // If this routine returns false, it is OK to call it again.  This allows
4351 // the user-defined signal handler to perform checks either before or after
4352 // the VM performs its own checks.  Naturally, the user code would be making
4353 // a serious error if it tried to handle an exception (such as a null check
4354 // or breakpoint) that the VM was generating for its own correct operation.
4355 //
4356 // This routine may recognize any of the following kinds of signals:
4357 //    SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4358 // It should be consulted by handlers for any of those signals.
4359 //
4360 // The caller of this routine must pass in the three arguments supplied
4361 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4362 // field of the structure passed to sigaction().  This routine assumes that
4363 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4364 //
4365 // Note that the VM will print warnings if it detects conflicting signal
4366 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4367 //
4368 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4369                                                  siginfo_t* siginfo,
4370                                                  void* ucontext,
4371                                                  int abort_if_unrecognized);
4372 
4373 void signalHandler(int sig, siginfo_t* info, void* uc) {
4374   assert(info != NULL && uc != NULL, "it must be old kernel");
4375   int orig_errno = errno;  // Preserve errno value over signal handler.
4376   JVM_handle_linux_signal(sig, info, uc, true);
4377   errno = orig_errno;
4378 }
4379 
4380 
4381 // This boolean allows users to forward their own non-matching signals
4382 // to JVM_handle_linux_signal, harmlessly.
4383 bool os::Linux::signal_handlers_are_installed = false;
4384 
4385 // For signal-chaining
4386 struct sigaction sigact[NSIG];
4387 uint64_t sigs = 0;
4388 #if (64 < NSIG-1)
4389 #error "Not all signals can be encoded in sigs. Adapt its type!"
4390 #endif
4391 bool os::Linux::libjsig_is_loaded = false;
4392 typedef struct sigaction *(*get_signal_t)(int);
4393 get_signal_t os::Linux::get_signal_action = NULL;
4394 
4395 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4396   struct sigaction *actp = NULL;
4397 
4398   if (libjsig_is_loaded) {
4399     // Retrieve the old signal handler from libjsig
4400     actp = (*get_signal_action)(sig);
4401   }
4402   if (actp == NULL) {
4403     // Retrieve the preinstalled signal handler from jvm
4404     actp = get_preinstalled_handler(sig);
4405   }
4406 
4407   return actp;
4408 }
4409 
4410 static bool call_chained_handler(struct sigaction *actp, int sig,
4411                                  siginfo_t *siginfo, void *context) {
4412   // Call the old signal handler
4413   if (actp->sa_handler == SIG_DFL) {
4414     // It's more reasonable to let jvm treat it as an unexpected exception
4415     // instead of taking the default action.
4416     return false;
4417   } else if (actp->sa_handler != SIG_IGN) {
4418     if ((actp->sa_flags & SA_NODEFER) == 0) {
4419       // automaticlly block the signal
4420       sigaddset(&(actp->sa_mask), sig);
4421     }
4422 
4423     sa_handler_t hand = NULL;
4424     sa_sigaction_t sa = NULL;
4425     bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4426     // retrieve the chained handler
4427     if (siginfo_flag_set) {
4428       sa = actp->sa_sigaction;
4429     } else {
4430       hand = actp->sa_handler;
4431     }
4432 
4433     if ((actp->sa_flags & SA_RESETHAND) != 0) {
4434       actp->sa_handler = SIG_DFL;
4435     }
4436 
4437     // try to honor the signal mask
4438     sigset_t oset;
4439     sigemptyset(&oset);
4440     pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4441 
4442     // call into the chained handler
4443     if (siginfo_flag_set) {
4444       (*sa)(sig, siginfo, context);
4445     } else {
4446       (*hand)(sig);
4447     }
4448 
4449     // restore the signal mask
4450     pthread_sigmask(SIG_SETMASK, &oset, NULL);
4451   }
4452   // Tell jvm's signal handler the signal is taken care of.
4453   return true;
4454 }
4455 
4456 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4457   bool chained = false;
4458   // signal-chaining
4459   if (UseSignalChaining) {
4460     struct sigaction *actp = get_chained_signal_action(sig);
4461     if (actp != NULL) {
4462       chained = call_chained_handler(actp, sig, siginfo, context);
4463     }
4464   }
4465   return chained;
4466 }
4467 
4468 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4469   if ((((uint64_t)1 << (sig-1)) & sigs) != 0) {
4470     return &sigact[sig];
4471   }
4472   return NULL;
4473 }
4474 
4475 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4476   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4477   sigact[sig] = oldAct;
4478   sigs |= (uint64_t)1 << (sig-1);
4479 }
4480 
4481 // for diagnostic
4482 int sigflags[NSIG];
4483 
4484 int os::Linux::get_our_sigflags(int sig) {
4485   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4486   return sigflags[sig];
4487 }
4488 
4489 void os::Linux::set_our_sigflags(int sig, int flags) {
4490   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4491   if (sig > 0 && sig < NSIG) {
4492     sigflags[sig] = flags;
4493   }
4494 }
4495 
4496 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4497   // Check for overwrite.
4498   struct sigaction oldAct;
4499   sigaction(sig, (struct sigaction*)NULL, &oldAct);
4500 
4501   void* oldhand = oldAct.sa_sigaction
4502                 ? CAST_FROM_FN_PTR(void*,  oldAct.sa_sigaction)
4503                 : CAST_FROM_FN_PTR(void*,  oldAct.sa_handler);
4504   if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4505       oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4506       oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4507     if (AllowUserSignalHandlers || !set_installed) {
4508       // Do not overwrite; user takes responsibility to forward to us.
4509       return;
4510     } else if (UseSignalChaining) {
4511       // save the old handler in jvm
4512       save_preinstalled_handler(sig, oldAct);
4513       // libjsig also interposes the sigaction() call below and saves the
4514       // old sigaction on it own.
4515     } else {
4516       fatal("Encountered unexpected pre-existing sigaction handler "
4517             "%#lx for signal %d.", (long)oldhand, sig);
4518     }
4519   }
4520 
4521   struct sigaction sigAct;
4522   sigfillset(&(sigAct.sa_mask));
4523   sigAct.sa_handler = SIG_DFL;
4524   if (!set_installed) {
4525     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4526   } else {
4527     sigAct.sa_sigaction = signalHandler;
4528     sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4529   }
4530   // Save flags, which are set by ours
4531   assert(sig > 0 && sig < NSIG, "vm signal out of expected range");
4532   sigflags[sig] = sigAct.sa_flags;
4533 
4534   int ret = sigaction(sig, &sigAct, &oldAct);
4535   assert(ret == 0, "check");
4536 
4537   void* oldhand2  = oldAct.sa_sigaction
4538                   ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4539                   : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4540   assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4541 }
4542 
4543 // install signal handlers for signals that HotSpot needs to
4544 // handle in order to support Java-level exception handling.
4545 
4546 void os::Linux::install_signal_handlers() {
4547   if (!signal_handlers_are_installed) {
4548     signal_handlers_are_installed = true;
4549 
4550     // signal-chaining
4551     typedef void (*signal_setting_t)();
4552     signal_setting_t begin_signal_setting = NULL;
4553     signal_setting_t end_signal_setting = NULL;
4554     begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4555                                           dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4556     if (begin_signal_setting != NULL) {
4557       end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4558                                           dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4559       get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4560                                          dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4561       libjsig_is_loaded = true;
4562       assert(UseSignalChaining, "should enable signal-chaining");
4563     }
4564     if (libjsig_is_loaded) {
4565       // Tell libjsig jvm is setting signal handlers
4566       (*begin_signal_setting)();
4567     }
4568 
4569     set_signal_handler(SIGSEGV, true);
4570     set_signal_handler(SIGPIPE, true);
4571     set_signal_handler(SIGBUS, true);
4572     set_signal_handler(SIGILL, true);
4573     set_signal_handler(SIGFPE, true);
4574 #if defined(PPC64)
4575     set_signal_handler(SIGTRAP, true);
4576 #endif
4577     set_signal_handler(SIGXFSZ, true);
4578 
4579     if (libjsig_is_loaded) {
4580       // Tell libjsig jvm finishes setting signal handlers
4581       (*end_signal_setting)();
4582     }
4583 
4584     // We don't activate signal checker if libjsig is in place, we trust ourselves
4585     // and if UserSignalHandler is installed all bets are off.
4586     // Log that signal checking is off only if -verbose:jni is specified.
4587     if (CheckJNICalls) {
4588       if (libjsig_is_loaded) {
4589         if (PrintJNIResolving) {
4590           tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4591         }
4592         check_signals = false;
4593       }
4594       if (AllowUserSignalHandlers) {
4595         if (PrintJNIResolving) {
4596           tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4597         }
4598         check_signals = false;
4599       }
4600     }
4601   }
4602 }
4603 
4604 // This is the fastest way to get thread cpu time on Linux.
4605 // Returns cpu time (user+sys) for any thread, not only for current.
4606 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4607 // It might work on 2.6.10+ with a special kernel/glibc patch.
4608 // For reference, please, see IEEE Std 1003.1-2004:
4609 //   http://www.unix.org/single_unix_specification
4610 
4611 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4612   struct timespec tp;
4613   int rc = os::Linux::clock_gettime(clockid, &tp);
4614   assert(rc == 0, "clock_gettime is expected to return 0 code");
4615 
4616   return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4617 }
4618 
4619 void os::Linux::initialize_os_info() {
4620   assert(_os_version == 0, "OS info already initialized");
4621 
4622   struct utsname _uname;
4623 
4624   uint32_t major;
4625   uint32_t minor;
4626   uint32_t fix;
4627 
4628   int rc;
4629 
4630   // Kernel version is unknown if
4631   // verification below fails.
4632   _os_version = 0x01000000;
4633 
4634   rc = uname(&_uname);
4635   if (rc != -1) {
4636 
4637     rc = sscanf(_uname.release,"%d.%d.%d", &major, &minor, &fix);
4638     if (rc == 3) {
4639 
4640       if (major < 256 && minor < 256 && fix < 256) {
4641         // Kernel version format is as expected,
4642         // set it overriding unknown state.
4643         _os_version = (major << 16) |
4644                       (minor << 8 ) |
4645                       (fix   << 0 ) ;
4646       }
4647     }
4648   }
4649 }
4650 
4651 uint32_t os::Linux::os_version() {
4652   assert(_os_version != 0, "not initialized");
4653   return _os_version & 0x00FFFFFF;
4654 }
4655 
4656 bool os::Linux::os_version_is_known() {
4657   assert(_os_version != 0, "not initialized");
4658   return _os_version & 0x01000000 ? false : true;
4659 }
4660 
4661 /////
4662 // glibc on Linux platform uses non-documented flag
4663 // to indicate, that some special sort of signal
4664 // trampoline is used.
4665 // We will never set this flag, and we should
4666 // ignore this flag in our diagnostic
4667 #ifdef SIGNIFICANT_SIGNAL_MASK
4668   #undef SIGNIFICANT_SIGNAL_MASK
4669 #endif
4670 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4671 
4672 static const char* get_signal_handler_name(address handler,
4673                                            char* buf, int buflen) {
4674   int offset = 0;
4675   bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4676   if (found) {
4677     // skip directory names
4678     const char *p1, *p2;
4679     p1 = buf;
4680     size_t len = strlen(os::file_separator());
4681     while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4682     jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4683   } else {
4684     jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4685   }
4686   return buf;
4687 }
4688 
4689 static void print_signal_handler(outputStream* st, int sig,
4690                                  char* buf, size_t buflen) {
4691   struct sigaction sa;
4692 
4693   sigaction(sig, NULL, &sa);
4694 
4695   // See comment for SIGNIFICANT_SIGNAL_MASK define
4696   sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4697 
4698   st->print("%s: ", os::exception_name(sig, buf, buflen));
4699 
4700   address handler = (sa.sa_flags & SA_SIGINFO)
4701     ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4702     : CAST_FROM_FN_PTR(address, sa.sa_handler);
4703 
4704   if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4705     st->print("SIG_DFL");
4706   } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4707     st->print("SIG_IGN");
4708   } else {
4709     st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4710   }
4711 
4712   st->print(", sa_mask[0]=");
4713   os::Posix::print_signal_set_short(st, &sa.sa_mask);
4714 
4715   address rh = VMError::get_resetted_sighandler(sig);
4716   // May be, handler was resetted by VMError?
4717   if (rh != NULL) {
4718     handler = rh;
4719     sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4720   }
4721 
4722   st->print(", sa_flags=");
4723   os::Posix::print_sa_flags(st, sa.sa_flags);
4724 
4725   // Check: is it our handler?
4726   if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4727       handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4728     // It is our signal handler
4729     // check for flags, reset system-used one!
4730     if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4731       st->print(
4732                 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4733                 os::Linux::get_our_sigflags(sig));
4734     }
4735   }
4736   st->cr();
4737 }
4738 
4739 
4740 #define DO_SIGNAL_CHECK(sig)                      \
4741   do {                                            \
4742     if (!sigismember(&check_signal_done, sig)) {  \
4743       os::Linux::check_signal_handler(sig);       \
4744     }                                             \
4745   } while (0)
4746 
4747 // This method is a periodic task to check for misbehaving JNI applications
4748 // under CheckJNI, we can add any periodic checks here
4749 
4750 void os::run_periodic_checks() {
4751   if (check_signals == false) return;
4752 
4753   // SEGV and BUS if overridden could potentially prevent
4754   // generation of hs*.log in the event of a crash, debugging
4755   // such a case can be very challenging, so we absolutely
4756   // check the following for a good measure:
4757   DO_SIGNAL_CHECK(SIGSEGV);
4758   DO_SIGNAL_CHECK(SIGILL);
4759   DO_SIGNAL_CHECK(SIGFPE);
4760   DO_SIGNAL_CHECK(SIGBUS);
4761   DO_SIGNAL_CHECK(SIGPIPE);
4762   DO_SIGNAL_CHECK(SIGXFSZ);
4763 #if defined(PPC64)
4764   DO_SIGNAL_CHECK(SIGTRAP);
4765 #endif
4766 
4767   // ReduceSignalUsage allows the user to override these handlers
4768   // see comments at the very top and jvm_md.h
4769   if (!ReduceSignalUsage) {
4770     DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4771     DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4772     DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4773     DO_SIGNAL_CHECK(BREAK_SIGNAL);
4774   }
4775 
4776   DO_SIGNAL_CHECK(SR_signum);
4777 }
4778 
4779 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4780 
4781 static os_sigaction_t os_sigaction = NULL;
4782 
4783 void os::Linux::check_signal_handler(int sig) {
4784   char buf[O_BUFLEN];
4785   address jvmHandler = NULL;
4786 
4787 
4788   struct sigaction act;
4789   if (os_sigaction == NULL) {
4790     // only trust the default sigaction, in case it has been interposed
4791     os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4792     if (os_sigaction == NULL) return;
4793   }
4794 
4795   os_sigaction(sig, (struct sigaction*)NULL, &act);
4796 
4797 
4798   act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4799 
4800   address thisHandler = (act.sa_flags & SA_SIGINFO)
4801     ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4802     : CAST_FROM_FN_PTR(address, act.sa_handler);
4803 
4804 
4805   switch (sig) {
4806   case SIGSEGV:
4807   case SIGBUS:
4808   case SIGFPE:
4809   case SIGPIPE:
4810   case SIGILL:
4811   case SIGXFSZ:
4812     jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4813     break;
4814 
4815   case SHUTDOWN1_SIGNAL:
4816   case SHUTDOWN2_SIGNAL:
4817   case SHUTDOWN3_SIGNAL:
4818   case BREAK_SIGNAL:
4819     jvmHandler = (address)user_handler();
4820     break;
4821 
4822   default:
4823     if (sig == SR_signum) {
4824       jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4825     } else {
4826       return;
4827     }
4828     break;
4829   }
4830 
4831   if (thisHandler != jvmHandler) {
4832     tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4833     tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4834     tty->print_cr("  found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4835     // No need to check this sig any longer
4836     sigaddset(&check_signal_done, sig);
4837     // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4838     if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4839       tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4840                     exception_name(sig, buf, O_BUFLEN));
4841     }
4842   } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4843     tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4844     tty->print("expected:");
4845     os::Posix::print_sa_flags(tty, os::Linux::get_our_sigflags(sig));
4846     tty->cr();
4847     tty->print("  found:");
4848     os::Posix::print_sa_flags(tty, act.sa_flags);
4849     tty->cr();
4850     // No need to check this sig any longer
4851     sigaddset(&check_signal_done, sig);
4852   }
4853 
4854   // Dump all the signal
4855   if (sigismember(&check_signal_done, sig)) {
4856     print_signal_handlers(tty, buf, O_BUFLEN);
4857   }
4858 }
4859 
4860 extern void report_error(char* file_name, int line_no, char* title,
4861                          char* format, ...);
4862 
4863 // this is called _before_ the most of global arguments have been parsed
4864 void os::init(void) {
4865   char dummy;   // used to get a guess on initial stack address
4866 //  first_hrtime = gethrtime();
4867 
4868   clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4869 
4870   init_random(1234567);
4871 
4872   Linux::set_page_size(sysconf(_SC_PAGESIZE));
4873   if (Linux::page_size() == -1) {
4874     fatal("os_linux.cpp: os::init: sysconf failed (%s)",
4875           os::strerror(errno));
4876   }
4877   init_page_sizes((size_t) Linux::page_size());
4878 
4879   Linux::initialize_system_info();
4880 
4881   Linux::initialize_os_info();
4882 
4883   // main_thread points to the aboriginal thread
4884   Linux::_main_thread = pthread_self();
4885 
4886   Linux::clock_init();
4887   initial_time_count = javaTimeNanos();
4888 
4889   // retrieve entry point for pthread_setname_np
4890   Linux::_pthread_setname_np =
4891     (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4892 
4893   os::Posix::init();
4894 }
4895 
4896 // To install functions for atexit system call
4897 extern "C" {
4898   static void perfMemory_exit_helper() {
4899     perfMemory_exit();
4900   }
4901 }
4902 
4903 void os::pd_init_container_support() {
4904   OSContainer::init();
4905 }
4906 
4907 // this is called _after_ the global arguments have been parsed
4908 jint os::init_2(void) {
4909 
4910   os::Posix::init_2();
4911 
4912   Linux::fast_thread_clock_init();
4913 
4914   // initialize suspend/resume support - must do this before signal_sets_init()
4915   if (SR_initialize() != 0) {
4916     perror("SR_initialize failed");
4917     return JNI_ERR;
4918   }
4919 
4920   Linux::signal_sets_init();
4921   Linux::install_signal_handlers();
4922 
4923   // Check and sets minimum stack sizes against command line options
4924   if (Posix::set_minimum_stack_sizes() == JNI_ERR) {
4925     return JNI_ERR;
4926   }
4927   Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4928 
4929 #if defined(IA32)
4930   workaround_expand_exec_shield_cs_limit();
4931 #endif
4932 
4933   Linux::libpthread_init();
4934   Linux::sched_getcpu_init();
4935   log_info(os)("HotSpot is running with %s, %s",
4936                Linux::glibc_version(), Linux::libpthread_version());
4937 
4938   if (UseNUMA) {
4939     if (!Linux::libnuma_init()) {
4940       UseNUMA = false;
4941     } else {
4942       if ((Linux::numa_max_node() < 1)) {
4943         // There's only one node(they start from 0), disable NUMA.
4944         UseNUMA = false;
4945       }
4946     }
4947     // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4948     // we can make the adaptive lgrp chunk resizing work. If the user specified
4949     // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4950     // disable adaptive resizing.
4951     if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4952       if (FLAG_IS_DEFAULT(UseNUMA)) {
4953         UseNUMA = false;
4954       } else {
4955         if (FLAG_IS_DEFAULT(UseLargePages) &&
4956             FLAG_IS_DEFAULT(UseSHM) &&
4957             FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4958           UseLargePages = false;
4959         } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4960           warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4961           UseAdaptiveSizePolicy = false;
4962           UseAdaptiveNUMAChunkSizing = false;
4963         }
4964       }
4965     }
4966     if (!UseNUMA && ForceNUMA) {
4967       UseNUMA = true;
4968     }
4969   }
4970 
4971   if (MaxFDLimit) {
4972     // set the number of file descriptors to max. print out error
4973     // if getrlimit/setrlimit fails but continue regardless.
4974     struct rlimit nbr_files;
4975     int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4976     if (status != 0) {
4977       log_info(os)("os::init_2 getrlimit failed: %s", os::strerror(errno));
4978     } else {
4979       nbr_files.rlim_cur = nbr_files.rlim_max;
4980       status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4981       if (status != 0) {
4982         log_info(os)("os::init_2 setrlimit failed: %s", os::strerror(errno));
4983       }
4984     }
4985   }
4986 
4987   // Initialize lock used to serialize thread creation (see os::create_thread)
4988   Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4989 
4990   // at-exit methods are called in the reverse order of their registration.
4991   // atexit functions are called on return from main or as a result of a
4992   // call to exit(3C). There can be only 32 of these functions registered
4993   // and atexit() does not set errno.
4994 
4995   if (PerfAllowAtExitRegistration) {
4996     // only register atexit functions if PerfAllowAtExitRegistration is set.
4997     // atexit functions can be delayed until process exit time, which
4998     // can be problematic for embedded VM situations. Embedded VMs should
4999     // call DestroyJavaVM() to assure that VM resources are released.
5000 
5001     // note: perfMemory_exit_helper atexit function may be removed in
5002     // the future if the appropriate cleanup code can be added to the
5003     // VM_Exit VMOperation's doit method.
5004     if (atexit(perfMemory_exit_helper) != 0) {
5005       warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5006     }
5007   }
5008 
5009   // initialize thread priority policy
5010   prio_init();
5011 
5012   return JNI_OK;
5013 }
5014 
5015 // Mark the polling page as unreadable
5016 void os::make_polling_page_unreadable(void) {
5017   if (!guard_memory((char*)_polling_page, Linux::page_size())) {
5018     fatal("Could not disable polling page");
5019   }
5020 }
5021 
5022 // Mark the polling page as readable
5023 void os::make_polling_page_readable(void) {
5024   if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5025     fatal("Could not enable polling page");
5026   }
5027 }
5028 
5029 // older glibc versions don't have this macro (which expands to
5030 // an optimized bit-counting function) so we have to roll our own
5031 #ifndef CPU_COUNT
5032 
5033 static int _cpu_count(const cpu_set_t* cpus) {
5034   int count = 0;
5035   // only look up to the number of configured processors
5036   for (int i = 0; i < os::processor_count(); i++) {
5037     if (CPU_ISSET(i, cpus)) {
5038       count++;
5039     }
5040   }
5041   return count;
5042 }
5043 
5044 #define CPU_COUNT(cpus) _cpu_count(cpus)
5045 
5046 #endif // CPU_COUNT
5047 
5048 // Get the current number of available processors for this process.
5049 // This value can change at any time during a process's lifetime.
5050 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5051 // If it appears there may be more than 1024 processors then we do a
5052 // dynamic check - see 6515172 for details.
5053 // If anything goes wrong we fallback to returning the number of online
5054 // processors - which can be greater than the number available to the process.
5055 int os::Linux::active_processor_count() {
5056   cpu_set_t cpus;  // can represent at most 1024 (CPU_SETSIZE) processors
5057   cpu_set_t* cpus_p = &cpus;
5058   int cpus_size = sizeof(cpu_set_t);
5059 
5060   int configured_cpus = os::processor_count();  // upper bound on available cpus
5061   int cpu_count = 0;
5062 
5063 // old build platforms may not support dynamic cpu sets
5064 #ifdef CPU_ALLOC
5065 
5066   // To enable easy testing of the dynamic path on different platforms we
5067   // introduce a diagnostic flag: UseCpuAllocPath
5068   if (configured_cpus >= CPU_SETSIZE || UseCpuAllocPath) {
5069     // kernel may use a mask bigger than cpu_set_t
5070     log_trace(os)("active_processor_count: using dynamic path %s"
5071                   "- configured processors: %d",
5072                   UseCpuAllocPath ? "(forced) " : "",
5073                   configured_cpus);
5074     cpus_p = CPU_ALLOC(configured_cpus);
5075     if (cpus_p != NULL) {
5076       cpus_size = CPU_ALLOC_SIZE(configured_cpus);
5077       // zero it just to be safe
5078       CPU_ZERO_S(cpus_size, cpus_p);
5079     }
5080     else {
5081        // failed to allocate so fallback to online cpus
5082        int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5083        log_trace(os)("active_processor_count: "
5084                      "CPU_ALLOC failed (%s) - using "
5085                      "online processor count: %d",
5086                      os::strerror(errno), online_cpus);
5087        return online_cpus;
5088     }
5089   }
5090   else {
5091     log_trace(os)("active_processor_count: using static path - configured processors: %d",
5092                   configured_cpus);
5093   }
5094 #else // CPU_ALLOC
5095 // these stubs won't be executed
5096 #define CPU_COUNT_S(size, cpus) -1
5097 #define CPU_FREE(cpus)
5098 
5099   log_trace(os)("active_processor_count: only static path available - configured processors: %d",
5100                 configured_cpus);
5101 #endif // CPU_ALLOC
5102 
5103   // pid 0 means the current thread - which we have to assume represents the process
5104   if (sched_getaffinity(0, cpus_size, cpus_p) == 0) {
5105     if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5106       cpu_count = CPU_COUNT_S(cpus_size, cpus_p);
5107     }
5108     else {
5109       cpu_count = CPU_COUNT(cpus_p);
5110     }
5111     log_trace(os)("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5112   }
5113   else {
5114     cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5115     warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5116             "which may exceed available processors", os::strerror(errno), cpu_count);
5117   }
5118 
5119   if (cpus_p != &cpus) { // can only be true when CPU_ALLOC used
5120     CPU_FREE(cpus_p);
5121   }
5122 
5123   assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5124   return cpu_count;
5125 }
5126 
5127 // Determine the active processor count from one of
5128 // three different sources:
5129 //
5130 // 1. User option -XX:ActiveProcessorCount
5131 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5132 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5133 //
5134 // Option 1, if specified, will always override.
5135 // If the cgroup subsystem is active and configured, we
5136 // will return the min of the cgroup and option 2 results.
5137 // This is required since tools, such as numactl, that
5138 // alter cpu affinity do not update cgroup subsystem
5139 // cpuset configuration files.
5140 int os::active_processor_count() {
5141   // User has overridden the number of active processors
5142   if (ActiveProcessorCount > 0) {
5143     log_trace(os)("active_processor_count: "
5144                   "active processor count set by user : %d",
5145                   ActiveProcessorCount);
5146     return ActiveProcessorCount;
5147   }
5148 
5149   int active_cpus;
5150   if (OSContainer::is_containerized()) {
5151     active_cpus = OSContainer::active_processor_count();
5152     log_trace(os)("active_processor_count: determined by OSContainer: %d",
5153                    active_cpus);
5154   } else {
5155     active_cpus = os::Linux::active_processor_count();
5156   }
5157 
5158   return active_cpus;
5159 }
5160 
5161 void os::set_native_thread_name(const char *name) {
5162   if (Linux::_pthread_setname_np) {
5163     char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5164     snprintf(buf, sizeof(buf), "%s", name);
5165     buf[sizeof(buf) - 1] = '\0';
5166     const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5167     // ERANGE should not happen; all other errors should just be ignored.
5168     assert(rc != ERANGE, "pthread_setname_np failed");
5169   }
5170 }
5171 
5172 bool os::distribute_processes(uint length, uint* distribution) {
5173   // Not yet implemented.
5174   return false;
5175 }
5176 
5177 bool os::bind_to_processor(uint processor_id) {
5178   // Not yet implemented.
5179   return false;
5180 }
5181 
5182 ///
5183 
5184 void os::SuspendedThreadTask::internal_do_task() {
5185   if (do_suspend(_thread->osthread())) {
5186     SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5187     do_task(context);
5188     do_resume(_thread->osthread());
5189   }
5190 }
5191 
5192 ////////////////////////////////////////////////////////////////////////////////
5193 // debug support
5194 
5195 bool os::find(address addr, outputStream* st) {
5196   Dl_info dlinfo;
5197   memset(&dlinfo, 0, sizeof(dlinfo));
5198   if (dladdr(addr, &dlinfo) != 0) {
5199     st->print(PTR_FORMAT ": ", p2i(addr));
5200     if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5201       st->print("%s+" PTR_FORMAT, dlinfo.dli_sname,
5202                 p2i(addr) - p2i(dlinfo.dli_saddr));
5203     } else if (dlinfo.dli_fbase != NULL) {
5204       st->print("<offset " PTR_FORMAT ">", p2i(addr) - p2i(dlinfo.dli_fbase));
5205     } else {
5206       st->print("<absolute address>");
5207     }
5208     if (dlinfo.dli_fname != NULL) {
5209       st->print(" in %s", dlinfo.dli_fname);
5210     }
5211     if (dlinfo.dli_fbase != NULL) {
5212       st->print(" at " PTR_FORMAT, p2i(dlinfo.dli_fbase));
5213     }
5214     st->cr();
5215 
5216     if (Verbose) {
5217       // decode some bytes around the PC
5218       address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5219       address end   = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5220       address       lowest = (address) dlinfo.dli_sname;
5221       if (!lowest)  lowest = (address) dlinfo.dli_fbase;
5222       if (begin < lowest)  begin = lowest;
5223       Dl_info dlinfo2;
5224       if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5225           && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5226         end = (address) dlinfo2.dli_saddr;
5227       }
5228       Disassembler::decode(begin, end, st);
5229     }
5230     return true;
5231   }
5232   return false;
5233 }
5234 
5235 ////////////////////////////////////////////////////////////////////////////////
5236 // misc
5237 
5238 // This does not do anything on Linux. This is basically a hook for being
5239 // able to use structured exception handling (thread-local exception filters)
5240 // on, e.g., Win32.
5241 void
5242 os::os_exception_wrapper(java_call_t f, JavaValue* value, const methodHandle& method,
5243                          JavaCallArguments* args, Thread* thread) {
5244   f(value, method, args, thread);
5245 }
5246 
5247 void os::print_statistics() {
5248 }
5249 
5250 bool os::message_box(const char* title, const char* message) {
5251   int i;
5252   fdStream err(defaultStream::error_fd());
5253   for (i = 0; i < 78; i++) err.print_raw("=");
5254   err.cr();
5255   err.print_raw_cr(title);
5256   for (i = 0; i < 78; i++) err.print_raw("-");
5257   err.cr();
5258   err.print_raw_cr(message);
5259   for (i = 0; i < 78; i++) err.print_raw("=");
5260   err.cr();
5261 
5262   char buf[16];
5263   // Prevent process from exiting upon "read error" without consuming all CPU
5264   while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5265 
5266   return buf[0] == 'y' || buf[0] == 'Y';
5267 }
5268 
5269 int os::stat(const char *path, struct stat *sbuf) {
5270   char pathbuf[MAX_PATH];
5271   if (strlen(path) > MAX_PATH - 1) {
5272     errno = ENAMETOOLONG;
5273     return -1;
5274   }
5275   os::native_path(strcpy(pathbuf, path));
5276   return ::stat(pathbuf, sbuf);
5277 }
5278 
5279 // Is a (classpath) directory empty?
5280 bool os::dir_is_empty(const char* path) {
5281   DIR *dir = NULL;
5282   struct dirent *ptr;
5283 
5284   dir = opendir(path);
5285   if (dir == NULL) return true;
5286 
5287   // Scan the directory
5288   bool result = true;
5289   char buf[sizeof(struct dirent) + MAX_PATH];
5290   while (result && (ptr = ::readdir(dir)) != NULL) {
5291     if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5292       result = false;
5293     }
5294   }
5295   closedir(dir);
5296   return result;
5297 }
5298 
5299 // This code originates from JDK's sysOpen and open64_w
5300 // from src/solaris/hpi/src/system_md.c
5301 
5302 int os::open(const char *path, int oflag, int mode) {
5303   if (strlen(path) > MAX_PATH - 1) {
5304     errno = ENAMETOOLONG;
5305     return -1;
5306   }
5307 
5308   // All file descriptors that are opened in the Java process and not
5309   // specifically destined for a subprocess should have the close-on-exec
5310   // flag set.  If we don't set it, then careless 3rd party native code
5311   // might fork and exec without closing all appropriate file descriptors
5312   // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5313   // turn might:
5314   //
5315   // - cause end-of-file to fail to be detected on some file
5316   //   descriptors, resulting in mysterious hangs, or
5317   //
5318   // - might cause an fopen in the subprocess to fail on a system
5319   //   suffering from bug 1085341.
5320   //
5321   // (Yes, the default setting of the close-on-exec flag is a Unix
5322   // design flaw)
5323   //
5324   // See:
5325   // 1085341: 32-bit stdio routines should support file descriptors >255
5326   // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5327   // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5328   //
5329   // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5330   // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5331   // because it saves a system call and removes a small window where the flag
5332   // is unset.  On ancient Linux kernels the O_CLOEXEC flag will be ignored
5333   // and we fall back to using FD_CLOEXEC (see below).
5334 #ifdef O_CLOEXEC
5335   oflag |= O_CLOEXEC;
5336 #endif
5337 
5338   int fd = ::open64(path, oflag, mode);
5339   if (fd == -1) return -1;
5340 
5341   //If the open succeeded, the file might still be a directory
5342   {
5343     struct stat64 buf64;
5344     int ret = ::fstat64(fd, &buf64);
5345     int st_mode = buf64.st_mode;
5346 
5347     if (ret != -1) {
5348       if ((st_mode & S_IFMT) == S_IFDIR) {
5349         errno = EISDIR;
5350         ::close(fd);
5351         return -1;
5352       }
5353     } else {
5354       ::close(fd);
5355       return -1;
5356     }
5357   }
5358 
5359 #ifdef FD_CLOEXEC
5360   // Validate that the use of the O_CLOEXEC flag on open above worked.
5361   // With recent kernels, we will perform this check exactly once.
5362   static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5363   if (!O_CLOEXEC_is_known_to_work) {
5364     int flags = ::fcntl(fd, F_GETFD);
5365     if (flags != -1) {
5366       if ((flags & FD_CLOEXEC) != 0)
5367         O_CLOEXEC_is_known_to_work = 1;
5368       else
5369         ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5370     }
5371   }
5372 #endif
5373 
5374   return fd;
5375 }
5376 
5377 
5378 // create binary file, rewriting existing file if required
5379 int os::create_binary_file(const char* path, bool rewrite_existing) {
5380   int oflags = O_WRONLY | O_CREAT;
5381   if (!rewrite_existing) {
5382     oflags |= O_EXCL;
5383   }
5384   return ::open64(path, oflags, S_IREAD | S_IWRITE);
5385 }
5386 
5387 // return current position of file pointer
5388 jlong os::current_file_offset(int fd) {
5389   return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5390 }
5391 
5392 // move file pointer to the specified offset
5393 jlong os::seek_to_file_offset(int fd, jlong offset) {
5394   return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5395 }
5396 
5397 // This code originates from JDK's sysAvailable
5398 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5399 
5400 int os::available(int fd, jlong *bytes) {
5401   jlong cur, end;
5402   int mode;
5403   struct stat64 buf64;
5404 
5405   if (::fstat64(fd, &buf64) >= 0) {
5406     mode = buf64.st_mode;
5407     if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5408       int n;
5409       if (::ioctl(fd, FIONREAD, &n) >= 0) {
5410         *bytes = n;
5411         return 1;
5412       }
5413     }
5414   }
5415   if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5416     return 0;
5417   } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5418     return 0;
5419   } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5420     return 0;
5421   }
5422   *bytes = end - cur;
5423   return 1;
5424 }
5425 
5426 // Map a block of memory.
5427 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5428                         char *addr, size_t bytes, bool read_only,
5429                         bool allow_exec) {
5430   int prot;
5431   int flags = MAP_PRIVATE;
5432 
5433   if (read_only) {
5434     prot = PROT_READ;
5435   } else {
5436     prot = PROT_READ | PROT_WRITE;
5437   }
5438 
5439   if (allow_exec) {
5440     prot |= PROT_EXEC;
5441   }
5442 
5443   if (addr != NULL) {
5444     flags |= MAP_FIXED;
5445   }
5446 
5447   char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5448                                      fd, file_offset);
5449   if (mapped_address == MAP_FAILED) {
5450     return NULL;
5451   }
5452   return mapped_address;
5453 }
5454 
5455 
5456 // Remap a block of memory.
5457 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5458                           char *addr, size_t bytes, bool read_only,
5459                           bool allow_exec) {
5460   // same as map_memory() on this OS
5461   return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5462                         allow_exec);
5463 }
5464 
5465 
5466 // Unmap a block of memory.
5467 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5468   return munmap(addr, bytes) == 0;
5469 }
5470 
5471 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5472 
5473 static clockid_t thread_cpu_clockid(Thread* thread) {
5474   pthread_t tid = thread->osthread()->pthread_id();
5475   clockid_t clockid;
5476 
5477   // Get thread clockid
5478   int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5479   assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5480   return clockid;
5481 }
5482 
5483 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5484 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5485 // of a thread.
5486 //
5487 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5488 // the fast estimate available on the platform.
5489 
5490 jlong os::current_thread_cpu_time() {
5491   if (os::Linux::supports_fast_thread_cpu_time()) {
5492     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5493   } else {
5494     // return user + sys since the cost is the same
5495     return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5496   }
5497 }
5498 
5499 jlong os::thread_cpu_time(Thread* thread) {
5500   // consistent with what current_thread_cpu_time() returns
5501   if (os::Linux::supports_fast_thread_cpu_time()) {
5502     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5503   } else {
5504     return slow_thread_cpu_time(thread, true /* user + sys */);
5505   }
5506 }
5507 
5508 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5509   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5510     return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5511   } else {
5512     return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5513   }
5514 }
5515 
5516 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5517   if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5518     return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5519   } else {
5520     return slow_thread_cpu_time(thread, user_sys_cpu_time);
5521   }
5522 }
5523 
5524 //  -1 on error.
5525 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5526   pid_t  tid = thread->osthread()->thread_id();
5527   char *s;
5528   char stat[2048];
5529   int statlen;
5530   char proc_name[64];
5531   int count;
5532   long sys_time, user_time;
5533   char cdummy;
5534   int idummy;
5535   long ldummy;
5536   FILE *fp;
5537 
5538   snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5539   fp = fopen(proc_name, "r");
5540   if (fp == NULL) return -1;
5541   statlen = fread(stat, 1, 2047, fp);
5542   stat[statlen] = '\0';
5543   fclose(fp);
5544 
5545   // Skip pid and the command string. Note that we could be dealing with
5546   // weird command names, e.g. user could decide to rename java launcher
5547   // to "java 1.4.2 :)", then the stat file would look like
5548   //                1234 (java 1.4.2 :)) R ... ...
5549   // We don't really need to know the command string, just find the last
5550   // occurrence of ")" and then start parsing from there. See bug 4726580.
5551   s = strrchr(stat, ')');
5552   if (s == NULL) return -1;
5553 
5554   // Skip blank chars
5555   do { s++; } while (s && isspace(*s));
5556 
5557   count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5558                  &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5559                  &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5560                  &user_time, &sys_time);
5561   if (count != 13) return -1;
5562   if (user_sys_cpu_time) {
5563     return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5564   } else {
5565     return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5566   }
5567 }
5568 
5569 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5570   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5571   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5572   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5573   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5574 }
5575 
5576 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5577   info_ptr->max_value = ALL_64_BITS;       // will not wrap in less than 64 bits
5578   info_ptr->may_skip_backward = false;     // elapsed time not wall time
5579   info_ptr->may_skip_forward = false;      // elapsed time not wall time
5580   info_ptr->kind = JVMTI_TIMER_TOTAL_CPU;  // user+system time is returned
5581 }
5582 
5583 bool os::is_thread_cpu_time_supported() {
5584   return true;
5585 }
5586 
5587 // System loadavg support.  Returns -1 if load average cannot be obtained.
5588 // Linux doesn't yet have a (official) notion of processor sets,
5589 // so just return the system wide load average.
5590 int os::loadavg(double loadavg[], int nelem) {
5591   return ::getloadavg(loadavg, nelem);
5592 }
5593 
5594 void os::pause() {
5595   char filename[MAX_PATH];
5596   if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5597     jio_snprintf(filename, MAX_PATH, "%s", PauseAtStartupFile);
5598   } else {
5599     jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5600   }
5601 
5602   int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5603   if (fd != -1) {
5604     struct stat buf;
5605     ::close(fd);
5606     while (::stat(filename, &buf) == 0) {
5607       (void)::poll(NULL, 0, 100);
5608     }
5609   } else {
5610     jio_fprintf(stderr,
5611                 "Could not open pause file '%s', continuing immediately.\n", filename);
5612   }
5613 }
5614 
5615 extern char** environ;
5616 
5617 // Run the specified command in a separate process. Return its exit value,
5618 // or -1 on failure (e.g. can't fork a new process).
5619 // Unlike system(), this function can be called from signal handler. It
5620 // doesn't block SIGINT et al.
5621 int os::fork_and_exec(char* cmd) {
5622   const char * argv[4] = {"sh", "-c", cmd, NULL};
5623 
5624   pid_t pid = fork();
5625 
5626   if (pid < 0) {
5627     // fork failed
5628     return -1;
5629 
5630   } else if (pid == 0) {
5631     // child process
5632 
5633     execve("/bin/sh", (char* const*)argv, environ);
5634 
5635     // execve failed
5636     _exit(-1);
5637 
5638   } else  {
5639     // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5640     // care about the actual exit code, for now.
5641 
5642     int status;
5643 
5644     // Wait for the child process to exit.  This returns immediately if
5645     // the child has already exited. */
5646     while (waitpid(pid, &status, 0) < 0) {
5647       switch (errno) {
5648       case ECHILD: return 0;
5649       case EINTR: break;
5650       default: return -1;
5651       }
5652     }
5653 
5654     if (WIFEXITED(status)) {
5655       // The child exited normally; get its exit code.
5656       return WEXITSTATUS(status);
5657     } else if (WIFSIGNALED(status)) {
5658       // The child exited because of a signal
5659       // The best value to return is 0x80 + signal number,
5660       // because that is what all Unix shells do, and because
5661       // it allows callers to distinguish between process exit and
5662       // process death by signal.
5663       return 0x80 + WTERMSIG(status);
5664     } else {
5665       // Unknown exit code; pass it through
5666       return status;
5667     }
5668   }
5669 }
5670 
5671 // is_headless_jre()
5672 //
5673 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5674 // in order to report if we are running in a headless jre
5675 //
5676 // Since JDK8 xawt/libmawt.so was moved into the same directory
5677 // as libawt.so, and renamed libawt_xawt.so
5678 //
5679 bool os::is_headless_jre() {
5680   struct stat statbuf;
5681   char buf[MAXPATHLEN];
5682   char libmawtpath[MAXPATHLEN];
5683   const char *xawtstr  = "/xawt/libmawt.so";
5684   const char *new_xawtstr = "/libawt_xawt.so";
5685   char *p;
5686 
5687   // Get path to libjvm.so
5688   os::jvm_path(buf, sizeof(buf));
5689 
5690   // Get rid of libjvm.so
5691   p = strrchr(buf, '/');
5692   if (p == NULL) {
5693     return false;
5694   } else {
5695     *p = '\0';
5696   }
5697 
5698   // Get rid of client or server
5699   p = strrchr(buf, '/');
5700   if (p == NULL) {
5701     return false;
5702   } else {
5703     *p = '\0';
5704   }
5705 
5706   // check xawt/libmawt.so
5707   strcpy(libmawtpath, buf);
5708   strcat(libmawtpath, xawtstr);
5709   if (::stat(libmawtpath, &statbuf) == 0) return false;
5710 
5711   // check libawt_xawt.so
5712   strcpy(libmawtpath, buf);
5713   strcat(libmawtpath, new_xawtstr);
5714   if (::stat(libmawtpath, &statbuf) == 0) return false;
5715 
5716   return true;
5717 }
5718 
5719 // Get the default path to the core file
5720 // Returns the length of the string
5721 int os::get_core_path(char* buffer, size_t bufferSize) {
5722   /*
5723    * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5724    * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5725    */
5726   const int core_pattern_len = 129;
5727   char core_pattern[core_pattern_len] = {0};
5728 
5729   int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5730   if (core_pattern_file == -1) {
5731     return -1;
5732   }
5733 
5734   ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5735   ::close(core_pattern_file);
5736   if (ret <= 0 || ret >= core_pattern_len || core_pattern[0] == '\n') {
5737     return -1;
5738   }
5739   if (core_pattern[ret-1] == '\n') {
5740     core_pattern[ret-1] = '\0';
5741   } else {
5742     core_pattern[ret] = '\0';
5743   }
5744 
5745   char *pid_pos = strstr(core_pattern, "%p");
5746   int written;
5747 
5748   if (core_pattern[0] == '/') {
5749     written = jio_snprintf(buffer, bufferSize, "%s", core_pattern);
5750   } else {
5751     char cwd[PATH_MAX];
5752 
5753     const char* p = get_current_directory(cwd, PATH_MAX);
5754     if (p == NULL) {
5755       return -1;
5756     }
5757 
5758     if (core_pattern[0] == '|') {
5759       written = jio_snprintf(buffer, bufferSize,
5760                              "\"%s\" (or dumping to %s/core.%d)",
5761                              &core_pattern[1], p, current_process_id());
5762     } else {
5763       written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
5764     }
5765   }
5766 
5767   if (written < 0) {
5768     return -1;
5769   }
5770 
5771   if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
5772     int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
5773 
5774     if (core_uses_pid_file != -1) {
5775       char core_uses_pid = 0;
5776       ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
5777       ::close(core_uses_pid_file);
5778 
5779       if (core_uses_pid == '1') {
5780         jio_snprintf(buffer + written, bufferSize - written,
5781                                           ".%d", current_process_id());
5782       }
5783     }
5784   }
5785 
5786   return strlen(buffer);
5787 }
5788 
5789 bool os::start_debugging(char *buf, int buflen) {
5790   int len = (int)strlen(buf);
5791   char *p = &buf[len];
5792 
5793   jio_snprintf(p, buflen-len,
5794                "\n\n"
5795                "Do you want to debug the problem?\n\n"
5796                "To debug, run 'gdb /proc/%d/exe %d'; then switch to thread " UINTX_FORMAT " (" INTPTR_FORMAT ")\n"
5797                "Enter 'yes' to launch gdb automatically (PATH must include gdb)\n"
5798                "Otherwise, press RETURN to abort...",
5799                os::current_process_id(), os::current_process_id(),
5800                os::current_thread_id(), os::current_thread_id());
5801 
5802   bool yes = os::message_box("Unexpected Error", buf);
5803 
5804   if (yes) {
5805     // yes, user asked VM to launch debugger
5806     jio_snprintf(buf, sizeof(char)*buflen, "gdb /proc/%d/exe %d",
5807                  os::current_process_id(), os::current_process_id());
5808 
5809     os::fork_and_exec(buf);
5810     yes = false;
5811   }
5812   return yes;
5813 }
5814 
5815 
5816 // Java/Compiler thread:
5817 //
5818 //   Low memory addresses
5819 // P0 +------------------------+
5820 //    |                        |\  Java thread created by VM does not have glibc
5821 //    |    glibc guard page    | - guard page, attached Java thread usually has
5822 //    |                        |/  1 glibc guard page.
5823 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5824 //    |                        |\
5825 //    |  HotSpot Guard Pages   | - red, yellow and reserved pages
5826 //    |                        |/
5827 //    +------------------------+ JavaThread::stack_reserved_zone_base()
5828 //    |                        |\
5829 //    |      Normal Stack      | -
5830 //    |                        |/
5831 // P2 +------------------------+ Thread::stack_base()
5832 //
5833 // Non-Java thread:
5834 //
5835 //   Low memory addresses
5836 // P0 +------------------------+
5837 //    |                        |\
5838 //    |  glibc guard page      | - usually 1 page
5839 //    |                        |/
5840 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()
5841 //    |                        |\
5842 //    |      Normal Stack      | -
5843 //    |                        |/
5844 // P2 +------------------------+ Thread::stack_base()
5845 //
5846 // ** P1 (aka bottom) and size (P2 = P1 - size) are the address and stack size
5847 //    returned from pthread_attr_getstack().
5848 // ** Due to NPTL implementation error, linux takes the glibc guard page out
5849 //    of the stack size given in pthread_attr. We work around this for
5850 //    threads created by the VM. (We adapt bottom to be P1 and size accordingly.)
5851 //
5852 #ifndef ZERO
5853 static void current_stack_region(address * bottom, size_t * size) {
5854   if (os::Linux::is_initial_thread()) {
5855     // initial thread needs special handling because pthread_getattr_np()
5856     // may return bogus value.
5857     *bottom = os::Linux::initial_thread_stack_bottom();
5858     *size   = os::Linux::initial_thread_stack_size();
5859   } else {
5860     pthread_attr_t attr;
5861 
5862     int rslt = pthread_getattr_np(pthread_self(), &attr);
5863 
5864     // JVM needs to know exact stack location, abort if it fails
5865     if (rslt != 0) {
5866       if (rslt == ENOMEM) {
5867         vm_exit_out_of_memory(0, OOM_MMAP_ERROR, "pthread_getattr_np");
5868       } else {
5869         fatal("pthread_getattr_np failed with error = %d", rslt);
5870       }
5871     }
5872 
5873     if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {
5874       fatal("Cannot locate current stack attributes!");
5875     }
5876 
5877     // Work around NPTL stack guard error.
5878     size_t guard_size = 0;
5879     rslt = pthread_attr_getguardsize(&attr, &guard_size);
5880     if (rslt != 0) {
5881       fatal("pthread_attr_getguardsize failed with error = %d", rslt);
5882     }
5883     *bottom += guard_size;
5884     *size   -= guard_size;
5885 
5886     pthread_attr_destroy(&attr);
5887 
5888   }
5889   assert(os::current_stack_pointer() >= *bottom &&
5890          os::current_stack_pointer() < *bottom + *size, "just checking");
5891 }
5892 
5893 address os::current_stack_base() {
5894   address bottom;
5895   size_t size;
5896   current_stack_region(&bottom, &size);
5897   return (bottom + size);
5898 }
5899 
5900 size_t os::current_stack_size() {
5901   // This stack size includes the usable stack and HotSpot guard pages
5902   // (for the threads that have Hotspot guard pages).
5903   address bottom;
5904   size_t size;
5905   current_stack_region(&bottom, &size);
5906   return size;
5907 }
5908 #endif
5909 
5910 static inline struct timespec get_mtime(const char* filename) {
5911   struct stat st;
5912   int ret = os::stat(filename, &st);
5913   assert(ret == 0, "failed to stat() file '%s': %s", filename, strerror(errno));
5914   return st.st_mtim;
5915 }
5916 
5917 int os::compare_file_modified_times(const char* file1, const char* file2) {
5918   struct timespec filetime1 = get_mtime(file1);
5919   struct timespec filetime2 = get_mtime(file2);
5920   int diff = filetime1.tv_sec - filetime2.tv_sec;
5921   if (diff == 0) {
5922     return filetime1.tv_nsec - filetime2.tv_nsec;
5923   }
5924   return diff;
5925 }
5926 
5927 /////////////// Unit tests ///////////////
5928 
5929 #ifndef PRODUCT
5930 
5931 #define test_log(...)              \
5932   do {                             \
5933     if (VerboseInternalVMTests) {  \
5934       tty->print_cr(__VA_ARGS__);  \
5935       tty->flush();                \
5936     }                              \
5937   } while (false)
5938 
5939 class TestReserveMemorySpecial : AllStatic {
5940  public:
5941   static void small_page_write(void* addr, size_t size) {
5942     size_t page_size = os::vm_page_size();
5943 
5944     char* end = (char*)addr + size;
5945     for (char* p = (char*)addr; p < end; p += page_size) {
5946       *p = 1;
5947     }
5948   }
5949 
5950   static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
5951     if (!UseHugeTLBFS) {
5952       return;
5953     }
5954 
5955     test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
5956 
5957     char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
5958 
5959     if (addr != NULL) {
5960       small_page_write(addr, size);
5961 
5962       os::Linux::release_memory_special_huge_tlbfs(addr, size);
5963     }
5964   }
5965 
5966   static void test_reserve_memory_special_huge_tlbfs_only() {
5967     if (!UseHugeTLBFS) {
5968       return;
5969     }
5970 
5971     size_t lp = os::large_page_size();
5972 
5973     for (size_t size = lp; size <= lp * 10; size += lp) {
5974       test_reserve_memory_special_huge_tlbfs_only(size);
5975     }
5976   }
5977 
5978   static void test_reserve_memory_special_huge_tlbfs_mixed() {
5979     size_t lp = os::large_page_size();
5980     size_t ag = os::vm_allocation_granularity();
5981 
5982     // sizes to test
5983     const size_t sizes[] = {
5984       lp, lp + ag, lp + lp / 2, lp * 2,
5985       lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
5986       lp * 10, lp * 10 + lp / 2
5987     };
5988     const int num_sizes = sizeof(sizes) / sizeof(size_t);
5989 
5990     // For each size/alignment combination, we test three scenarios:
5991     // 1) with req_addr == NULL
5992     // 2) with a non-null req_addr at which we expect to successfully allocate
5993     // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
5994     //    expect the allocation to either fail or to ignore req_addr
5995 
5996     // Pre-allocate two areas; they shall be as large as the largest allocation
5997     //  and aligned to the largest alignment we will be testing.
5998     const size_t mapping_size = sizes[num_sizes - 1] * 2;
5999     char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6000       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6001       -1, 0);
6002     assert(mapping1 != MAP_FAILED, "should work");
6003 
6004     char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6005       PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6006       -1, 0);
6007     assert(mapping2 != MAP_FAILED, "should work");
6008 
6009     // Unmap the first mapping, but leave the second mapping intact: the first
6010     // mapping will serve as a value for a "good" req_addr (case 2). The second
6011     // mapping, still intact, as "bad" req_addr (case 3).
6012     ::munmap(mapping1, mapping_size);
6013 
6014     // Case 1
6015     test_log("%s, req_addr NULL:", __FUNCTION__);
6016     test_log("size            align           result");
6017 
6018     for (int i = 0; i < num_sizes; i++) {
6019       const size_t size = sizes[i];
6020       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6021         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6022         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " ->  " PTR_FORMAT " %s",
6023                  size, alignment, p2i(p), (p != NULL ? "" : "(failed)"));
6024         if (p != NULL) {
6025           assert(is_aligned(p, alignment), "must be");
6026           small_page_write(p, size);
6027           os::Linux::release_memory_special_huge_tlbfs(p, size);
6028         }
6029       }
6030     }
6031 
6032     // Case 2
6033     test_log("%s, req_addr non-NULL:", __FUNCTION__);
6034     test_log("size            align           req_addr         result");
6035 
6036     for (int i = 0; i < num_sizes; i++) {
6037       const size_t size = sizes[i];
6038       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6039         char* const req_addr = align_up(mapping1, alignment);
6040         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6041         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6042                  size, alignment, p2i(req_addr), p2i(p),
6043                  ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6044         if (p != NULL) {
6045           assert(p == req_addr, "must be");
6046           small_page_write(p, size);
6047           os::Linux::release_memory_special_huge_tlbfs(p, size);
6048         }
6049       }
6050     }
6051 
6052     // Case 3
6053     test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6054     test_log("size            align           req_addr         result");
6055 
6056     for (int i = 0; i < num_sizes; i++) {
6057       const size_t size = sizes[i];
6058       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6059         char* const req_addr = align_up(mapping2, alignment);
6060         char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6061         test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " ->  " PTR_FORMAT " %s",
6062                  size, alignment, p2i(req_addr), p2i(p), ((p != NULL ? "" : "(failed)")));
6063         // as the area around req_addr contains already existing mappings, the API should always
6064         // return NULL (as per contract, it cannot return another address)
6065         assert(p == NULL, "must be");
6066       }
6067     }
6068 
6069     ::munmap(mapping2, mapping_size);
6070 
6071   }
6072 
6073   static void test_reserve_memory_special_huge_tlbfs() {
6074     if (!UseHugeTLBFS) {
6075       return;
6076     }
6077 
6078     test_reserve_memory_special_huge_tlbfs_only();
6079     test_reserve_memory_special_huge_tlbfs_mixed();
6080   }
6081 
6082   static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6083     if (!UseSHM) {
6084       return;
6085     }
6086 
6087     test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6088 
6089     char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6090 
6091     if (addr != NULL) {
6092       assert(is_aligned(addr, alignment), "Check");
6093       assert(is_aligned(addr, os::large_page_size()), "Check");
6094 
6095       small_page_write(addr, size);
6096 
6097       os::Linux::release_memory_special_shm(addr, size);
6098     }
6099   }
6100 
6101   static void test_reserve_memory_special_shm() {
6102     size_t lp = os::large_page_size();
6103     size_t ag = os::vm_allocation_granularity();
6104 
6105     for (size_t size = ag; size < lp * 3; size += ag) {
6106       for (size_t alignment = ag; is_aligned(size, alignment); alignment *= 2) {
6107         test_reserve_memory_special_shm(size, alignment);
6108       }
6109     }
6110   }
6111 
6112   static void test() {
6113     test_reserve_memory_special_huge_tlbfs();
6114     test_reserve_memory_special_shm();
6115   }
6116 };
6117 
6118 void TestReserveMemorySpecial_test() {
6119   TestReserveMemorySpecial::test();
6120 }
6121 
6122 #endif